A View on Drug Development for Cancer Prevention

Cancer-related morbidity and mortality can be significantly reduced by the successful implementation of cancer prevention measures (1, 2). These measures may include lifestyle modification or other interventions to reduce cancer risk, but this article is dedicated specifically to the development of medicines to prevent cancer. This is a topic with a long history (3, 4) and has been known by various names, including chemoprophylaxis (5, 6), chemoprevention (7), and cancer interception (8). Historically, chemoprevention was further subdivided into primary chemoprevention (medicines to prevent cancer in healthy and/or high-risk individuals), secondary chemoprevention (medicines for the treatment of premalignant lesions), or tertiary chemoprevention (medicines for the prevention of cancer recurrence; refs. 9, 10). This article focuses mainly on primary and secondary chemoprevention.

The premise of chemoprevention is sometimes based on the concept that many cancers are preceded by an identifiable phase of “premalignant disease” and that, during this phase, a medicine might be used to prevent the transition to malignant disease; it can also be based on the concept that cancer develops in individuals with specific risk factors where, again, it may be appropriate to intervene with a medicine to prevent cancer (4, 8, 9, 11–13). There are many different examples of premalignant disease and many different examples of specific risk factors. An example of the former is preinvasive precursor lesions, such as colonic polyps, which are precursor lesions to colorectal cancer (14, 15). An example of the latter is people with genetic predisposition to cancer, such as women with a germline BRCA1 mutation (gBRCA1m) or BRCA2 mutation (gBRCA2m) who are at increased risk of breast and ovarian cancers, also known as hereditary breast and ovarian cancer (HBOC) syndrome (16, 17). For further specific examples of premalignant disease and cancer risk factors, see Fig. 1A–F and Supplementary Table S1.

Medicines developed for cancer prevention may deliver a spectrum of benefits, the foremost being to avoid the morbi­dity and mortality associated with a cancer diagnosis and cancer treatment. However, medicines might also bring clinical benefit by replacing the surgical procedures (or ablative therapies) used to treat early-stage preinvasive lesions (e.g., colonic polyps). Medicines might also serve as an alternative to prophylactic risk-reducing surgeries that are currently recommended to reduce cancer risk, such as bilateral mastectomy for HBOC syndrome. In addition, we postulate that medicines have a greater chance of being effective (and achieving cure) when used during the premalignant disease phase, at least in part because tumor heterogeneity, tumor burden, and cancer's ability to evade the immune response (all significant drivers of cancer treatment failure) tend to increase as cancer progresses (18–20). There have already been notable successes in the development of medicines to prevent cancer, including human papillomavirus (HPV) vaccines for cervical cancer prevention (21). However, progress in drug development for cancer prevention has, in general, lagged behind the development of medicines for advanced disease. This is most likely because, despite the enormous opportunity to prevent disease, there are also considerable challenges to address and barriers to overcome when developing medicines for cancer prevention (4, 9, 11, 12, 22–28).

Here we review the agents that have been successfully developed for cancer prevention thus far, and we discuss some of the agents that were less successful. We further highlight other therapies that are either currently being tested or might be tested in the future. We provide a perspective on the opportunities and challenges for drug development in cancer prevention, including a series of concepts that may enable or accelerate the development of medicines for cancer prevention in the future.

Several agents have been successfully developed for cancer prevention. To illustrate this point, here we highlight several FDA-approved agents with a relevant label (Table 1). The agents are discussed in the context of specific disease areas, namely, breast cancer, esophageal cancer (Barrett esophagus), bladder cancer, skin cancer (actinic keratosis), and cancers associated with HPV or hepatitis B virus (HBV) infection.

Endocrine therapy was repurposed for testing in the cancer prevention setting following prior success in the treatment of advanced-stage breast cancer (29, 30). The first large prevention study was the pivotal NSABP-P1 randomized control trial (RCT; ref. 31). Women at increased risk of breast cancer were randomized to 5 years of treatment with tamoxifen (n = 6,681) or placebo (n = 6,707). Women were eligible if they (i) were ≥60 years old, (ii) were 35–59 years old with ≥1.66% risk of breast cancer over the next 5 years (Gail model), or (iii) had a history of lobular carcinoma in situ (LCIS). Importantly, the risk of individuals developing breast cancer, which was the primary outcome measure in the trial, was roughly halved in women who took tamoxifen (31). In 1998, this led to the FDA approval of tamoxifen as a treatment to reduce the incidence of breast cancer in (pre- and postmenopausal) women at high risk (Table 1).

Raloxifene was originally developed for the prevention, or treatment, of osteoporosis in postmenopausal women, for which it was approved in 1997 and 1999, respectively. In 2007, raloxifene was also licensed to reduce the risk of invasive breast cancer in postmenopausal women at high risk for invasive breast cancer (Table 1) based on the results of four large RCTs: MORE (32), CORE (33), and RUTH (34), which all demonstrated a reduced risk of invasive breast cancer in women treated with raloxifene compared with placebo, and STAR (35), which demonstrated that raloxifene is as effective as tamoxifen in reducing the risk of invasive breast cancer.

However, a recent meta-analysis showed that endocrine therapy to reduce breast cancer risk is taken by only 16.3% of eligible women (36). The reasons underlying this low uptake are complex and are both patient-related and physician-related (37–39). Some women are deterred by the relatively common adverse effects of endocrine therapy (hot flushes and sexual dysfunction) and, although rare, the increased risk of more severe adverse effects (thromboembolic events and endometrial cancer; refs. 37–39). This illustrates that, despite demonstrable efficacy in cancer prevention, there can be barriers to medicine uptake in eligible individuals. In the future, toxicity issues might be surmounted by administering endocrine therapy at a lower dose, or via an alternative administration route, or through use of newer endocrine therapies with a superior safety profile (38, 39). Other work shows that uptake is greater in women who have higher breast cancer risk, such as genetic predisposition or a prior diagnosis of ductal carcinoma in situ (DCIS) or LCIS (40–42), which may emphasize the importance of targeting cancer prevention strategies at those who are at the very highest risk of cancer.

Photodynamic therapy (PDT) in combination with the photosensitizer drug porfimer sodium (PDT/PS) is licensed for the treatment of Barrett esophagus (BE) with high-grade dysplasia (HGD; Table 1). PDT/PS was first approved for the treatment of advanced-stage esophageal cancer in 1995 (43). Further to this, studies examining PDT/PS for earlier-stage disease reported that it could eliminate BE in some patients (44, 45). In a pivotal RCT, Overholt and colleagues (46) randomized patients with BE and HGD to PDT/PS (n = 138) or omeprazole (n = 70). Complete ablation of HGD was superior with PDT/PS compared with omeprazole (77% vs. 39%), as was the proportion of BE replaced by normal epithelium (52% vs. 7%), and fewer patients progressed to cancer with PDT/PS compared with omeprazole (13% vs. 28%; ref. 46). These and other data were pivotal in the FDA approval for PDT/PS for BE with HGD granted in 2003 (46). Although further data confirmed the efficacy of PDT/PS for BE (47), its use has now been superseded by other effective therapies for the treatment of BE, including endoscopic resection and radiofrequency ablation, mainly due to the side effects of PDT such as increased photosensitivity (43). However, this BE example is important, not least because it shows that agents can play a role in the prevention of cancer by treating individuals with an identifiable high-risk preinvasive lesion.

Intravesical administration of Bacillus Calmette–Guerin (BCG), a live attenuated strain of Mycobacterium bovis, is licensed for the treatment of carcinoma in situ (CIS) of the bladder (Table 1). BCG was originally developed as a tuberculosis vaccine in the 1920s, but subsequent observations suggested the potential of BCG as an immunotherapy-based treatment for localized cancer (48–50). In 1976, Morales conducted a landmark study in nine patients with recurrent non–muscle invasive bladder cancer (NMIBC), demonstrating that intravesical BCG reduced bladder cancer recurrence (51). This was then followed by three RCTs that confirmed the activity of BCG in NMIBC (52–54), including a study in which 131 patients with CIS were randomized to doxorubicin or BCG, which demonstrated a superior rate of complete response for BCG versus doxorubicin (70% vs. 34%; ref. 54), data that were pivotal in the FDA approval that was granted in 1990 (49).

There are cases in which disease is either unresponsive to BCG therapy or recurs despite BCG therapy. In 1998, the FDA approved intravesical use of valrubicin for the treatment of BCG-unresponsive CIS based on a single-arm study involving 90 patients in which 21% achieved a complete response (55, 56). In 2020, the FDA approved pembrolizumab for the treatment of BCG-unresponsive CIS based on a single-arm study involving 96 patients in which 41% achieved a complete response (57). Current National Comprehensive Cancer Network (NCCN) Guidelines for bladder cancer (V3.2022) recommend intravesical BCG as an induction or maintenance therapy to prevent disease recurrence following transurethral resection (TUR) in patients with CIS of the bladder, while valrubicin and pembrolizumab are listed as therapeutic options for BCG-refractory or BCG-unresponsive CIS (https://www.nccn.org/guidelines/). The approval of both BCG and pembrolizumab for CIS of the bladder demonstrates that immunotherapy can play a role in the treatment of preinvasive lesions.

Actinic keratosis is a chronic skin condition characterized by localized scaly patches on the skin (caused by long-term sun exposure), which may progress to cutaneous squamous cell carcinoma (cSCC; ref. 58). PDT combined with 5-aminolevulinic acid is FDA approved for the treatment of actinic keratosis, as are several topical agents, including fluorouracil cream, diclofenac gel, imiquimod cream, ingenol mebutate gel, and tirbanibulin ointment; the condition can also be managed with cryotherapy or surgery (refs. 59, 60; Table 1). With a range of effective therapies now available, treatment selection is individually tailored based on both the patient and lesion characteristics (60). Efficacy, measured in terms of lesion clearance, has been typically used to support regulatory approvals for actinic keratosis (59, 61). The most recent approval in 2020, for tirbanibulin ointment, was based on two RCTs in which a total of 702 patients were randomized to tirbanibulin ointment or vehicle. The primary outcome measure was complete lesion clearance, which was achieved in 49% of the patients in the tirbanibulin group versus 9% in the vehicle group (62). Actinic keratosis is therefore another example demonstrating that medicines can be developed successfully to treat individuals with a premalignant lesion.

The causal link between HPV infection and cancers of the cervix, anogenital tract, and oropharynx has paved the way for HPV vaccines in cancer prevention (63–66). Based on the data obtained in appropriate RCTs, three HPV vaccines have now been FDA approved for the prevention of HPV-associated cancers (Table 1; ref. 21). Before the first RCTs were launched, in 2001 the FDA recommended that RCTs should be powered to demonstrate efficacy in preventing high-grade cervical lesions [cervical intraepithelial neoplasia (CIN) stage 2 (CIN2) or stage 3 (CIN3)], a position that was also later endorsed by the World Health Organization (WHO) in 2004 (67). Thus, in a pivotal RCT of the quadrivalent vaccine, young women were randomized to vaccine or placebo with the primary endpoint being the prevention of cervical intraepithelial neoplasia (CIN2 or CIN3), adenocarcinoma in situ, or cervical cancer associated with HPV-16 or -18 (68). In a pivotal phase III study of the bivalent vaccine, the primary endpoint was the prevention of CIN2 associated with HPV-16 or -18 (69). Importantly, for both vaccines, efficacy in HPV-naive females was greater than 90%, but in females with a prior or active HPV infection, efficacy was significantly reduced (68, 69). These data, as well as further studies demonstrating efficacy and safety in younger populations, meant that subsequent vaccination programs have been targeted at adolescents predominantly (21, 64). Postmarketing commitments for these vaccines included the sponsors providing further longer-term safety and efficacy data, and further published studies have since confirmed the efficacy and safety, as well as the durability of the protection afforded (21).

In 2020, the WHO published a global strategy to eliminate cervical cancer as a public health problem, which is based on all countries meeting a minimum set of targets for vaccination, screening, and treatment (70). National HPV vaccination programs for the prevention of HPV-associated cancers are now active across the world, with immunization targeted mainly at young females, but also young males in some territories due to evidence that HPV vaccination of males is also beneficial (21, 71). The implementation of HPV vaccination in conjunction with cervical cancer screening and treatment has radically reduced both the global incidence of cervical cancer and death from cervical cancer (72–75). However, limitations in access to vaccination and screening in low-income and lower-middle-income countries must be surmounted in order to reach the goal of eliminating cervical cancer as a global public health problem (74).

Around 80% of hepatocellular carcinoma (HCC) diagnoses are associated with viral hepatitis caused by HBV infection, and HBV infection is the predominant cause of HCC in many Asian countries where HBV is endemic (76). In 1983, Taiwan implemented a universal HBV vaccination program to prevent HBV-associated viral hepatitis and HCC. Long-term follow-up of vaccinated and unvaccinated individuals in Taiwan shows that HBV vaccination protects against the development of HCC in both children and young adults (77, 78). The FDA has approved three HBV vaccines for the prevention of HBV infection (Table 1), and the WHO now recommends universal HBV vaccination for infants. This policy has been adopted by many countries and may be driving a global decrease in the incidence of HBV-related HCC (79). The examples of HPV and HBV vaccination demonstrate the considerable potential of vaccination as a drug modality in cancer prevention.

Here we discuss (i) agents that have been tested, but that did not achieve product labeling, (ii) agents that are the subject of ongoing studies, and (iii) agents that might be tested in the future. Due to the myriad of agents investigated already or currently under investigation across multiple tumor types (12, 28, 39, 80), only selected examples are presented here.

Some of the earliest trials of medicines for cancer prevention focused on dietary supplements, such as vitamins and antioxidants (3). Although based on strong evidence that diet influences cancer risk, to date, the results of these studies have been disappointing, either due to lack of efficacy or toxicity (81).

Two large studies for lung cancer prevention and a third for prostate cancer prevention are important examples that failed to demonstrate efficacy. The α-tocopherol (vitamin E), β-carotene prevention (ATBC) study randomized 29,133 male smokers to one of four regimens: α-tocopherol alone, β-carotene alone, both agents combined, or placebo (82). None of the treatment arms reduced lung cancer incidence—in fact a higher lung cancer incidence was seen in those receiving β-carotene, which was associated with heavier smoking and higher alcohol intake (82, 83). Another study, CARET, which randomized 18,314 high-risk men (smokers, former smokers, or asbestos exposure) to β-carotene plus vitamin A versus placebo, was closed early because of higher incidence of lung cancer, and higher mortality related to both lung cancer and cardiovascular disease (CVD), in the intervention arm (84). The SELECT study, an RCT in 35,533 healthy men, failed to demonstrate efficacy for selenium and vitamin E, alone or in combination, for the prevention of prostate cancer (85). Retinoic acid (a synthetic derivative of vitamin A) is an example that showed excessive toxicity. Although studies in oral leukoplakia (86) and cervical dysplasia (87) demonstrated significant lesion regression, retinoids were not approved because of the numerous toxicities associated with long-term administration (88). These findings have tempered enthusiasm for cancer prevention approaches based on dietary supplements.

The role of chronic inflammation in cancer initiation suggests that anti-inflammatory drugs may have the potential for cancer prevention (89). Thus far, the majority of supporting data come from studies of nonsteroidal anti-inflammatory drugs (NSAID), including aspirin, celecoxib, and sulindac.

Evidence that NSAIDs could reduce polyp burden, colorectal cancer incidence, and colorectal cancer mortality first emerged in the 1980s and 1990s (90–92), triggering a considerable research effort to examine NSAIDs in colorectal cancer prevention, including observational studies and prospective RCTs (93). Meta-analyses of these studies show that aspirin usage is associated with a 20% to 30% reduction in colorectal cancer risk (91, 94). The strength of the evidence led the U.S. Preventive Services Task Force to first recommend aspirin for colorectal cancer prevention in 2007 (95), with updated guidelines in 2016 recommending low-dose aspirin (81 mg daily) for colorectal cancer and CVD prevention for those ages 50 to 59 years with a 10% 10-year risk of CVD, while for individuals 60 to 69 years, the decision to initiate aspirin should be based on an individual assessment of benefits and harms; recommendations for those <50 years or ≥70 years could not be made due to insufficient data (96).

There is also work examining NSAIDs to prevent cancer for individuals with hereditary colon cancer (28). Most recently, long-term follow-up of the CaPP2 study, which randomized 861 participants with Lynch syndrome to aspirin (600 mg daily) or placebo, demonstrated a significantly reduced risk of colorectal cancer in those taking aspirin (97). In 2000, these and other data led the UK's National Institute for Health and Care Excellence to recommend daily aspirin, taken for more than 2 years, to reduce the risk of colorectal cancer in people with Lynch syndrome (https://www.nice.org.uk/guidance/ng151).

However, there are still outstanding questions regarding the use of aspirin for cancer prevention, including selection of the appropriate dose, optimal duration of treatment, suitability for the elderly, and whether biomarkers can be utilized to individualize aspirin use for disease prevention (93, 94). For example, while most data suggest that regular (at least two times per week), long-term (5–10 years) use of aspirin is required to see significant benefit, the optimal age at which an individual should begin taking aspirin is unclear (94), and more recent data, linking long-term aspirin use with detrimental outcomes in the elderly (≥70 years of age; ref. 98), have left uncertainties regarding use in the elderly (99). Further epidemiologic evidence suggests aspirin usage also reduces the risk of other gastrointestinal cancers (including stomach, pancreas, and esophageal; refs. 100, 101), although we are not aware of published guidelines recommending aspirin use as prevention for these cancers. Future studies of aspirin for cancer prevention will provide further data and will hopefully address many of these outstanding questions.

In the early 1990s, evidence suggesting that NSAIDs target cancer through inhibition of COX-2, and concerns over bleeding risk with long-term aspirin use, led to the development of highly selective COX-2 inhibitors, including celecoxib (15, 102, 103). Steinbach and colleagues (104) randomized 77 patients with familial adenomatous polyposis (FAP) to treatment with celecoxib or placebo. Patients underwent endoscopy to monitor polyp burden at baseline and after 6 months of celecoxib treatment. Polyp burden was significantly reduced in the celecoxib arm compared with the placebo arm. The drug was well tolerated, with no significant differences in adverse events observed between the celecoxib and placebo arms (104). Based on these data, in 1999 the FDA granted accelerated approval for celecoxib to reduce the number of adenomatous colorectal polyps in FAP as an adjunct to usual care (15). However, while two further RCTs confirmed that celecoxib was efficacious in reducing polyp burden (105, 106), several large studies demonstrated that COX-2 inhibitors were associated with an increased risk of CVD events (106–109), which led to the manufacturer withdrawing colorectal cancer prevention from the celecoxib drug label in 2011. This highlights the need for long-term monitoring of safety data beyond drug approval, as it may reveal adverse effects that were not initially evident.

Combining NSAIDs with other agents is also being explored for cancer prevention. Following on from promising results obtained in animal models combining NSAIDs with the polyamine synthesis inhibitor difluoromethylornithine (DFMO), in 2008 Meyskens and colleagues (110) reported the results of an RCT in which 375 individuals with a history of resected adenomas were randomized to receive DFMO plus sulindac versus placebo. The combination was associated with an impressive 70% decrease in the recurrence of all adenomas with minimal toxicity (110). A larger study is ongoing and will provide more data on both the efficacy and safety of this combination for the prevention of adenomas and colorectal cancer (NCT01349881).

One of the largest RCTs published to date testing aspirin for cancer prevention was the AspECT trial (111), which tested the proton pump inhibitor (PPI) esomeprazole with or without aspirin for patients with BE. Prior to this study, there was conflicting evidence and debate regarding the benefit of these drugs as prophylaxis for BE (111). In AspECT, 2,557 patients with BE were randomized to receive a high or low dose of esomeprazole with or without aspirin for at least 8 years. A composite primary endpoint of time to all-cause mortality, esophageal adenocarcinoma, or HGD was used. The study concluded that the high-dose PPI and aspirin combination did improve outcomes in patients with BE and was safe (111). However, as discussed in detail elsewhere (112, 113), due to several caveats of the study, there is an open debate as to whether combination therapy with PPIs and aspirin should be offered to patients with BE.

Other anti-inflammatory drugs, beyond NSAIDs, may have the potential for cancer prevention. Interleukin-1β is a proinflammatory cytokine with a role in autoimmune disease, CVD, and cancer, while canakinumab is an interleukin-1β inhibitory antibody that is already licensed to treat a range of immune-related disorders (114). CANTOS was a large RCT that demonstrated that canakinumab can reduce CVD in patients with previous myocardial infarction (115). An additional analysis of CANTOS revealed that canakinumab significantly reduced both lung cancer incidence and mortality (116). On the basis of these data, four further trials of canakinumab in non–small cell lung cancer (NSCLC) were launched by the sponsor: three phase III (CANOPY-1 and CANOPY-2 in locally advanced or metastatic NSCLC; CANOPY-A in adjuvant NSCLC) and a phase II study (CANOPY-N in neoadjuvant NSCLC). The results of CANOPY-1 and CANOPY-2 were reported recently: Both RCTs were unsuccessful with neither trial meeting the primary efficacy endpoints (117). One potential explanation is that proinflammatory interleukin-1β signaling is more important in the biology of early disease than metastatic disease. This might be addressed in CANOPY-A and CANOPY-N, which have yet to report (117). We are not aware of any studies evaluating interleukin-1β inhibition prospectively in a primary chemoprevention setting, although such studies would appear warranted.

The 5α-reductase enzymes catalyze the synthesis of dihydrotesterone, a hormone involved in prostate development and growth, and prostate cancer. Finasteride is a 5α-reductase type 1 inhibitor, whereas dutasteride inhibits 5α-reductase types 1 and 2. Both agents had been previously approved for the treatment of benign prostatic hyperplasia before being tested for prostate cancer prevention (118). In the Prostate Cancer Prevention Trial (119), in the final analysis population 9,060 men ≥55 years of age with prostate-specific antigen (PSA) level ≤3 ng/mL and a normal digital rectal examination (DRE) were randomized to finasteride or placebo. In the REDUCE trial (120), in the final analysis population 6,729 men ages 50 to 75 with PSA 2.5 to 10 ng/mL and with a prior negative prostate biopsy were randomized to dutasteride or placebo. In both RCTs, men receiving the 5α reductase inhibitor had a statistically significant reduction in the incidence of prostate cancer of around 25% (finasteride) or 23% (dutasteride; refs. 119, 120), but there were issues common to both trials that prevented either drug being approved (12, 121, 122). First, the results were driven in part by an end-of-study biopsy, in addition to biopsies that were triggered by PSA or DRE, which may have limited the generalizability of the results to general practice. Secondly, in both trials, the overall cancer risk reduction was driven entirely by a reduction in Gleason ≤6 tumors (that meet the criteria for clinically insignificant disease and are unlikely to lead to prostate cancer mortality). Third, in both studies, there was a paradoxical increase in the incidence of high-grade tumors in men receiving 5α-reductase inhibitor, which is, arguably, the most significant concern that precluded approval (12, 121, 122). Despite further work demonstrating that increased detection of high-grade tumors in patients receiving a 5α-reductase inhibitor may be an artifact, and long-term follow-up data demonstrating a reduction in prostate cancer incidence in men receiving these agents, 5α-reductase inhibitors are unlikely to be approved for cancer prevention due to lingering doubts regarding their suitability for that purpose (80, 121, 123).

Cancer risk due to genetic predisposition may be either monogenic (due to the presence of a single moderate-risk or high-risk allele) or polygenic (multiple genetic variants each contribute a small effect to risk). Most of the inherited cancer syndromes can be considered monogenic, because cancer risk is conferred by the presence of a single moderate-risk or high-risk allele (124). In cancer syndromes, all cells in the body are heterozygous for the mutation that can be considered the first hit required for tumorigenesis. Functional loss of the second gene copy and further somatic mutations are then required for cancer development (124). Some of the cancer syndromes are caused by mutations in oncogenes that are often druggable (124). For example, hereditary papillary renal cell carcinoma (HPRCC) is driven by MET mutations, and because small-molecule inhibitors of the MET receptor have activity in MET-mutant kidney cancers (125), in theory, these same drugs might be tested for cancer prevention in HPRCC. However, most of the known cancer syndromes involve loss of function in tumor suppressor genes (124), including HBOC (loss of BRCA1 or BRCA2 gene function), von Hippel–Lindau (VHL) syndrome (loss of VHL gene function), and Lynch syndrome [characterized by mismatch repair (MMR) deficiency]. Although loss of function in tumor suppressor genes makes drug development more challenging, there are emerging opportunities for cancer prevention.

For example, evidence linking the RANK/RANKL signaling pathway with BRCA1m breast cancer, and work showing that denosumab (a RANKL inhibitory antibody) could suppress tumor progression in BRCA1m preclinical models (126), led to the BRCA-P RCT, which is testing whether denosumab can prevent breast cancer in gBRCA1m women (NCT04711109). Another potential approach to prevent HBOC-associated cancers could be PARP inhibitors, which have been licensed to treat breast, ovarian, pancreatic, and prostate cancers associated with a BRCA1 or BRCA2 mutation based on the principle of synthetic lethality (127). Although we are not aware of any clinical trials testing PARP inhibitors for cancer prevention, preclinical data are supportive of the approach (128, 129). In patients with VHL syndrome, loss of VHL gene function is associated with deregulated expression of the HIF2α protein and the development of tumors in several organs, including the kidney and central nervous system. Belzutifan is a HIF2α inhibitor that has been successfully developed to treat cancers associated with VHL syndrome (130). In theory, HIF2α inhibitors might also be tested to prevent cancer in VHL syndrome. Lynch syndrome is defined by the presence of pathogenic germline mutations in one of several DNA MMR genes, and individuals are at increased risk of developing MMR-deficient (MMRd) colorectal cancer and endometrial cancer (124). In common with spontaneous MMRd cancers, the MMRd colorectal tumors that arise in patients with Lynch syndrome are typically highly responsive to immunotherapy (131). As discussed later, this sensitivity might also be leveraged as a prevention strategy for Lynch syndrome.

Another potential prevention approach for cancer syndromes is to deploy CRISPR–Cas9-based gene editing to correct the genetic defect in the tissues deemed to be at risk. Recently, CRISPR–Cas9-based therapy has shown promise for the treatment of other monogenic (nononcology) disorders, such as transthyretin amyloidosis (132). However, challenges remain to be surmounted for the use of CRISPR–Cas9 in cancer therapy, including the effective delivery of the therapy to all target cells (133).

Several lines of evidence suggest that immune-checkpoint inhibitors (ICI) may be effective in the treatment of premalignant disease. First, BCG and pembrolizumab have already been approved for the treatment of CIS of the bladder (49, 57). Second, impressive results, in terms of pathologic response, have been obtained with ICIs in the neoadjuvant treatment of MMRd colorectal cancer, melanoma, and NSCLC (134–137), which supports the use of ICIs in earlier-stage disease. Third, profiling of preinvasive lesions indicates that they must evade immune destruction to progress and that this is associated with the upregulation of immune checkpoints (20).

Several ongoing studies are testing ICIs for the treatment of premalignant disease beyond CIS of the bladder. Building on data suggesting that immune evasion occurs prior to invasion in lung squamous cell carcinoma (138–140), one study is randomizing 42 participants with bronchial dysplasia to nivolumab or placebo, with improvement in lesion histology as the primary endpoint (NCT03347838). Lung adenocarcinomas may also develop from preinvasive precursor lesions [which present as pulmonary nodules (PN)] through a process that involves evasion of immunity (141). Therefore, another study is evaluating pembrolizumab with regression of PNs as the primary endpoint (NCT03634241). Profiling of oral premalignant lesions, which are precursors to oral cancer, also suggests a role for immune evasion in lesion progression (142–146), while preclinical studies showed that ICIs could suppress the progression of oral premalignant lesions in a carcinogen-induced model (147–151). Four studies are therefore examining ICIs for oral premalignant lesions (NCT02882282, NCT03603223, NCT03692325, and NCT04504552). Building on the activity of ICIs in MMRd cancers (131), ICIs are also being tested for the prevention of cancer in Lynch syndrome (NCT03631641 and NCT04711434). Other studies are examining ICIs for CIN (NCT04712851) and DCIS (NCT02872025).

Mature data are awaited from these studies to understand if ICIs will be effective and safe for these settings. Meanwhile, the field is exploring an array of additional approaches to target the immune response in advanced-stage disease, including bispecific ICIs and immune cell engagers (152), cell therapies (153), and myeloid cell targets (154). It will be important to consider the potential of these new immunotherapy approaches for cancer prevention in the future.

The role of HPV and HBV vaccination in cancer prevention highlights the potential to target other pathogens or microbes for cancer prevention. At least 11 pathogens are recognized as being causative of human cancer, including seven types of virus, three species of platyhelminthes, and one bacterial species (Fig. 1; refs. 155, 156). The bacterial species, Helicobacter pylori, is causative in both gastric adenocarcinoma and mucosa-associated lymphoid tissue lymphoma (157). Studies have examined the effect of Helicobacter pylori eradication therapy (typically a PPI combined with two antibiotics). A meta-analysis of RCTs showed that eradication therapy reduces both incidence of gastric cancer and gastric cancer–related mortality (158). However, as yet, no country has implemented a national screening and treatment program for Helicobacter pylori.

More recent work has implicated other bacterial species in the carcinogenesis of several tumor types (156, 159) and that the composition of the microbiome influences cancer development, response to cancer therapies, and treatment-related toxicities (156, 159, 160). For example, detrimental alterations in the gut microbiome (gut dysbiosis) have been linked with the development of colorectal cancer, and a role for specific microbial species (e.g., Fusobacterium nucleatum) in colorectal cancer carcinogenesis has been identified (159, 161). Several studies showed that gut microbiome composition influenced the response to ICIs, and that the ICI responder and nonresponder phenotypes seen in humans could be recapitulated in mice using fecal microbiota transplantation (FMT; refs. 162–165). Clinical trials of microbiome-based medicines for the treatment of advanced-stage cancer are currently underway (156, 159).

Future prospective studies might explore the targeting of specific pathogens, or the broader microbiome, to prevent cancer. As well as the more traditional modalities (e.g., vaccines, antibiotics, or antiviral drugs), novel microbiome-based medicines, such as those designed to alter the gut microbiome composition (e.g., FMT, dietary supplements, prebiotics, probiotics; refs. 156, 159), might also be explored for cancer prevention.

The development of vaccines to prevent nonpathogen-associated cancers has been proposed (166, 167). A major obstacle is that this requires robust prediction of the antigens that will occur in a tumor before it develops (168). For many cancers, the mutations occurring are highly variable (which makes predicting tumor antigens difficult), and/or the resulting antigens are poorly immunogenic. There may be exceptions where the approach is more feasible, such as Lynch syndrome, in which the associated MMRd tumors produce a relatively large number of predictable frameshift peptide neoantigens that are both immunogenic and shared between patients (168). Preclinical and early-phase clinical studies are exploring several different vaccine approaches for MMRd cancers, and promising initial data on safety, immunogenicity, and efficacy have been obtained (168). However, further studies are required to fully establish the potential of vaccines for the prevention of cancers in Lynch syndrome, as well as other tumor types.

Other work is addressing whether vaccines can treat advanced-stage cancers by targeting tumor neoantigens. Thus far, progress has been slow, with several candidate vaccines failing to demonstrate significant efficacy in phase III studies (169). However, recent advances, including a greater understanding of mechanisms that compromise vaccine efficacy, new vaccine delivery platforms (such as RNA vaccines), and the potential to combine vaccines with other therapies (such as ICIs) may accelerate our ability to generate more effective cancer vaccine approaches for both advanced-stage cancers and cancer prevention (169–171).

Further research has identified other pathways that are aberrant in premalignant disease that might be targeted for cancer prevention (124, 172–179). For example, EGFR mutations have been detected in lung adenocarcinoma in situ, in which they may represent a targetable oncogenic driver event (176, 180). Mutation of the tumor suppressor gene TP53 (a downstream consequence of which is increased genomic instability) occurs in 80% of serous tubal intraepithelial carcinomas, which are a precursor lesion to ovarian cancer (177). And finally, the most common somatic mutations observed in clonal hematopoiesis are present in genes associated with epigenetic modification, including TET2, DNMT3A, and ASXL1 (178). Further therapeutic strategies for prevention might target specific hallmark traits of cancer, such as deregulated cellular metabolism or senescent cells (160). Additional relevant targets will likely be identified in the PreCancer Atlas initiative, which aims to characterize the genetic, epigenetic, proteomic, and microenvironmental landscape of premalignant tumors (181, 182). The ability to effectively drug these targets may be facilitated by recent advances in medicinal chemistry (183, 184). For example, in terms of targets where it is desirable to block either expression or activity, this may be facilitated by new modalities such as oligonucleotide-based therapies (which suppress the synthesis of the protein; refs. 183, 185) or protein-degrading molecules (that target the protein for proteolytic degradation; refs. 183, 186).

As described above, several medicines have been successfully developed for cancer prevention based on clinical data demonstrating both efficacy and a favorable safety profile in the target population. Other studies of medicines for cancer prevention have been less successful, typically due to lack of efficacy or excessive toxicity. Furthermore, there is continued interest in developing medicines for cancer prevention, with a range of drug modalities being tested currently, while other agents may be developed in the future.

However, there are several challenges that arise when developing drugs for cancer prevention, including identifying the appropriate high-risk population that would benefit from a medicine; selecting the right medicine (in terms of both the right target and safety profile); and choosing the appropriate dose, schedule, and duration of therapy. There are further challenges associated with running successful trials, including picking the appropriate primary outcome measure, and successfully recruiting otherwise healthy individuals to a cancer prevention trial. Additional barriers that may preclude progress include several disincentives for drug developers to enter the cancer prevention space, the cost-effectiveness of medicines developed for cancer prevention, and lack of awareness regarding the importance of early detection and early treatment.

Strategies that can potentially overcome these challenges are likely to increase the chances of success in the future. To develop these strategies, not only can we draw from insights obtained in previous cancer prevention research, but we can also take inspiration from advances in drug development in other disease areas. Furthermore, we can think innovatively about how to maximize success; this includes, for example, harnessing new technologies. Next, we synthesize this thinking into some generalized concepts that may either enable or accelerate the development of medicines for cancer prevention.

Concepts that may enable or accelerate drug development for cancer prevention are organized here under six themes: unmet clinical need, patient identification, risk stratification, pharmacological intervention, clinical trials, and health care policy.

The first theme focuses on identifying scenarios in which medicines could meet an important unmet clinical need. Although this will be driven by a desire to improve on patient outcomes, we also note the importance of carefully weighing the risks, benefits, and financial costs of the proposed medication versus the same for the existing standard of care. Furthermore, we highlight the importance of understanding the views of patients before embarking on the development of a medicine for cancer prevention.

Specific examples of where we believe there is a high unmet need for medicines in a cancer-preventative setting are described below. The first example we highlight is the development of medicines for the treatment of premalignant solid lesions, especially where surgical resection or ablative therapy is either difficult, suboptimal, or associated with significant morbidity. One example is when multiple lesions may be present due to “field cancerization” (e.g., bronchial dysplasia or oral premalignant lesions), in which it is difficult to completely resect or ablate all lesions (187). The second example we highlight is the development of medicines for cancer prevention in settings where the first line of treatment for advanced disease is a surgical procedure associated with significant morbidity, such as cystectomy or radical prostatectomy. The third example is the development of medicines for premalignant lesions that are prone to recur and are currently managed by surveillance and repeated rounds of surgery. One example is the management of colonic polyps, where the chronic administration of a drug that suppresses polyp formation (i) may be preferable to repeated rounds of endoscopic polypectomy and (ii) could help prevent the incidence of interval cancers (28, 188). The fourth example is the development of medicines to tackle the premalignant precursors to hematologic malignancies, for example, myelodysplastic syndrome (MDS) or smoldering multiple myeloma (sMM; refs. 189, 190), in which a systemically administered medicine may offer the only possibility to prevent progression to advanced-stage hematologic malignancy. The fifth example is the development of medicines to prevent cancer for individuals at high risk due to genetic predisposition (13, 17, 28, 124) or due to an underlying condition, such as liver cirrhosis (191–193). The final example we highlight is the development of medicines to prevent cancers in which the survival outcomes are still very poor. For all stages combined, the 5-year relative survival rate is still very low for cancers of the pancreas (11%), liver (20%), esophagus (20%), and lung (22%; ref. 194). These are just some examples, and other settings with high unmet clinical needs can also be envisioned.

The current standard of care for cancer syndromes (prior to the development of any cancer) typically involves surveillance and sometimes prophylactic surgery to reduce cancer risk. If individuals with a cancer syndrome do develop cancer, then standard treatments for cancer may be used, including surgery, radiotherapy, and medicines. There are significant impacts in managing cancer syndromes in this way, including the burden on health care systems of managing a chronic lifelong condition, the psychologic and quality of life (QoL) impacts on affected individuals, and the side effects associated with surgery and treatments for advanced-stage cancer (195). Risk-reducing surgical procedures, such as bilateral mastectomy and/or salpingo-oophorectomy (for HBOC; ref. 39), total abdominal colectomy (for Lynch syndrome; ref. 28), or total gastrectomy (for HDGC; ref. 196), are associated with significant QoL impacts. For patients with a cancer syndrome, medicines developed for cancer prevention could be impactful in several ways, including (i) delaying or replacing the need for prophylactic surgery; (ii) reducing the burden of surveillance, either in terms of frequency of visits or duration of surveillance necessary; and (iii) preventing the development of advanced-stage disease and the associated cancer treatment, morbidity, and mortality.

Any medicine developed for cancer prevention should be compared with the existing standard of care, with the benefits, risks, and financial cost of a medicine being weighed against the same for the existing standard of care. For example, individuals diagnosed with a specific premalignant lesion might already be offered surgery as a standard of care, and the benefits, risks, and costs of a medicine will need to be weighed against the same for the standard-of-care surgical approach. Understanding patient preference is also important. For example, if a medicine were available as an alternative to surgery, some individuals may prefer surgery, while others may prefer a medicine. The decision may be dependent on the risks and benefits associated with each approach. In this case, we propose that drug development for cancer prevention might not always be driven by a need to completely replace the existing standard of care; it may instead be motivated by the need to provide patients with a different option that is equally effective and safe. To address patient-centric issues associated with cancer prevention strategies, engagement with patient advocates and patient surveys can be undertaken before embarking on the development of a medicine (197, 198).

For several cancer types, population-based screening already identifies patients with premalignant disease. Two prominent examples include DCIS detected by mammography (199) or colorectal polyps detected by colonoscopy (200). However, there are many cancers for which early detection is still suboptimal, such as ovarian cancer and pancreatic cancer, to name two examples. With increasing focus on early cancer detection as a significant research priority (201–203), we may expect to see a stage shift: more cancers diagnosed at an early stage, more premalignant disease diagnoses, and more individuals identified as being at high risk of cancer. An array of methodologies and technologies for identifying early cancers, premalignant disease, or those at risk of cancer have been developed, or are being developed, which may either be applied in a population-based fashion or targeted to specific segments of the population (202). In this subsection, we discuss some of these technologies and their potential impact.

We will focus here on three broad categories of technology: (i) blood-based liquid biopsy, which uses a blood sample to detect cancer; (ii) the analysis of biological samples other than blood to detect cancer; and (iii) digital technologies to detect cancer (Fig. 2; Supplementary Table S2).

Tumor markers in blood, for example, CA125 (for ovarian) or PSA (for prostate), may already be utilized in clinical practice for screening and follow-up of patients, although their utility for early cancer detection has limitations (204, 205). However, the liquid biopsy field is examining a broader range of blood-based analytes for early detection (Fig. 2). One example is the detection of circulating tumor DNA (ctDNA) or circulating cell-free DNA (cfDNA), which is based on the principle that cancers release specific DNA species into the blood that can facilitate early cancer detection (206, 207). Although the sensitivity of this approach is tumor type– and disease stage–dependent, it may be well suited to detect certain cancers very early. For example, several studies indicate that this approach may be particularly well suited to detect early-stage liver cancers, and at least two tests utilizing ctDNA and/or cfDNA for the early detection of HCC have been granted FDA breakthrough device designation (208). Other analytes explored in liquid biopsy include circulating extracellular vesicles (209, 210) and T-cell receptor repertoire profiling (211, 212).

Early detection is also being explored through the analysis of biological samples other than blood. This has a particularly strong rationale when the sample acquisition is proximal to the tissue at risk, a principle already used for colorectal cancer screening by the analysis of stool samples for occult blood (213). However, the field is also examining a broader range of sample types, including the analysis of volatile molecules in breath (breathomics), to detect lung cancer (214) or the analysis of cfDNA in fluids other than blood (215), such as urinary tumor DNA to detect bladder cancer (Fig. 2; ref. 216). A further example is Cytosponge-TFF3, which utilizes a minimally invasive device to sample esophageal epithelial cells to detect BE (217, 218). In a large RCT, it was shown that offering Cytosponge-TFF3 to high-risk individuals significantly improved the detection of BE and early cancer when compared with standard clinical practice, and the test had high patient acceptability (219).

Digital technology has for many years led to improvements in the ability to detect cancers radiologically. Further to this, machine learning algorithms can now be trained to detect cancerous cells in images with a performance that can match, or even exceed, a trained physician. Recent developments include algorithms to detect breast cancer in mammograms (220), lung cancer in CT scans (221), polyps during colonoscopy (222), and skin cancers in photographs of the skin (223). Advances in digital technology are also enabling new devices for early detection, including capsule endoscopy (a capsule containing a camera that can be swallowed to perform endoscopy of the gastrointestinal tract; refs. 224, 225) and implantable or wearable devices that can monitor human health in real time (226, 227).

One specific technology that may significantly impact early detection is the development of multiple cancer early detection (MCED) tests (228–234). These tests, which typically combine two or more early detection technologies, are designed to maximize sensitivity and specificity for detecting multiple cancer types in a single test format. For example, one MCED test with FDA breakthrough device designation combines methylated cfDNA detection in blood with machine learning to identify signatures indicative of specific cancers (231, 234), while another is based on the detection of both methylated cfDNA and protein biomarkers in the blood to identify specific cancers (228, 233). The diagnostic performance of MCED tests has been established in retrospective studies using blood obtained from thousands of subjects with and without cancer. Importantly, MCED tests have relatively high sensitivity to detect cancer types for which screening is currently not standard in average-risk individuals (including pancreatic and liver cancers) and have high specificity (which is important to reduce the incidence of false positives and the associated potential harms; refs. 228–234). Two prospective studies of MCED tests in people with no history of cancer have been reported, with both demonstrating that MCED tests can detect new cancers in asymptomatic individuals (233, 235). Further ongoing, prospective studies of MCED tests will provide additional information on feasibility, acceptability, safety, and impact on cancer outcomes (236–238).

It is currently unclear as to whether MCED tests will be able to detect premalignant disease. This is because the published data on the sensitivity of MCED tests focus mainly on the detection of invasive cancers (stage 1–4 disease) and not on the detection of premalignant disease. Furthermore, published work shows that the overall sensitivity of MCED tests for stage 1 disease is typically much lower when compared with later-stage disease. By extrapolation, due to this lower overall sensitivity for early-stage disease, it may be challenging to detect premalignant disease with the current generation of MCED tests. Further research will be necessary to understand whether an MCED test can be developed to detect premalignant disease.

How can we identify individuals with genetic predisposition to cancer in the general population? In the era of next-generation sequencing, testing for the high-risk alleles associated with known cancer syndromes is now efficient and cost-effective but is typically limited to high-risk individuals, such as patients who already have a diagnosis of cancer, or individuals with a family history of cancer (239). Unfortunately, many individuals who carry these high-risk alleles are undetected in the general population until the patient presents with symptomatic disease. Indeed, in the United Kingdom and United States, it is estimated that around 90% of asymptomatic BRCA1/2m carriers remain undetected (240, 241). However, published economic analyses have demonstrated that population-based screening for BRCA mutations is cost-effective when it is linked to effective management and treatment (e.g., surveillance and/or risk-reducing surgery as per guideline recommendations; ref. 242). There are strong arguments for population-based screening for moderate-to-high penetrance cancer predisposition genes, which have both established clinical utility and actionable management guidelines (such as BRCA mutations), and this may become more common in health care systems if the clinical impact and cost-effectiveness can be justified (239, 243).

Not all individuals who are identified as belonging to a high-risk group will progress to cancer. For example, while around 50% of bronchial CIS lesions progress, the other 50% undergo spontaneous regression or remain indolent (175, 244). Risk stratification entails the identification of the individuals who are at the highest risk of progression. The rationale for risk stratification in clinical trials, as well as some of the methods to achieve risk stratification, is discussed below.

Risk stratification is important in the context of drug development for cancer prevention for the following reasons: (i) We need to target medicines to patients who will significantly benefit, without exposing those who will not benefit to unnecessary treatment and associated potential toxicities; (ii) there will be more events in high-risk individuals (because they have a greater chance of progression to malignant cancer), and therefore RCTs of new medicines in high-risk populations can be smaller, faster, and less costly; and (iii) the risks associated with testing a new medicine may be more acceptable in a high-risk population because of greater potential for benefit. Furthermore, trials conducted in high-risk populations can provide an important proof of concept before moving into broader, lower-risk populations. In further support of this approach, during the clinical development of blood pressure– and low-density lipoprotein–modifying drugs (now widely used to prevent CVD), the earliest studies that demonstrated meaningful clinical benefit were conducted in higher-risk populations, which then paved the way for extending these therapies to broader, lower-risk populations (245).

Consequentially, trials of medicines designed to prevent cancer may often need to include an algorithm to select for higher-risk individuals. For some hematologic malignancies, established and routinely utilized algorithms are already available that identify individuals with low-, moderate-, or high-risk premalignant disease. For example, the revised International Prognostic Scoring System (IPSS-R) for MDS and the Mayo Clinic model for sMM are algorithms that stratify patients in terms of their risk of progression to acute myeloid leukemia or multiple myeloma, respectively (Supplementary Table S1; refs. 246, 247). In a recent RCT (248), 182 patients with mostly intermediate-risk or high-risk sMM were randomized to lenalidomide or observation. The primary endpoint was progression to symptomatic multiple myeloma. Within 36 months of randomization, 31.6% of patients in the observation arm had progressed versus 7.3% in the lenalidomide arm. Therefore, within a 3-year time frame, this relatively small randomized study demonstrated that lenalidomide significantly delayed the progression of sMM to advanced-stage disease (HR = 0.28). Furthermore, the greatest benefit (HR = 0.09) was achieved in patients with the highest-risk sMM, who have approximately 25% risk of progression to multiple myeloma per year (248). Of note, previous studies of other therapies in an unselected sMM patient population were unsuccessful, due to significant toxicity and without demonstrating significant clinical benefit (249–251). Although lenalidomide is unlikely to be approved for sMM, due to reasons that are discussed in detail elsewhere (252), the study clearly demonstrates the impact of risk stratification in a cancer prevention setting.

Models for predicting future cancer risk, which typically incorporate both genetic and nongenetic risk factors, have also been reported for some solid cancers, including breast, colorectal, kidney, and ovarian cancers (202, 253–258). Algorithms that predict the progression of specific preinvasive solid lesions are also in development. For example, the recently described CompCyst model for pancreatic cysts utilizes a combination of clinical features, imaging features, and cyst fluid markers to identify patients at risk of progression to pancreatic cancer (259). Other recent work has identified genetic, epigenetic, or immunologic features that predict the risk of progression for bronchial premalignant lesions (139, 140, 175) and BE (260, 261).

Another strategy to identifying individuals at higher risk is to focus on cancer syndromes. For example, individuals diagnosed with FAP have an almost 100% lifetime risk of developing colorectal cancer (Supplementary Table S1; ref. 28). For other cancer syndromes, while the overall lifetime risk of cancer is still substantially higher than the general population, a portion of mutation carriers never develop the associated cancer. In these cases, individuals might be further stratified utilizing additional risk factors, such as polygenic risk scores (PRS) or family history of cancer. A population-based PRS was strongly associated with the risk of breast and ovarian cancers in BRCA1/2 mutation carriers (262).

Many of the published models for predicting future cancer risk have been facilitated by studying large national or international cohorts of patients. Similar large cohorts of individuals might also be assembled to identify and validate disease risk modifiers in other high-risk populations, such as individuals with clonal hematopoiesis (263), individuals with nonalcoholic fatty liver disease (264), or individuals who are at risk due to other factors such as pathogen or carcinogen exposure. Moreover, if these cohorts were established prospectively (by recruiting individuals diagnosed in the health care system), and if recruitment included consent to reapproach individuals for participation in future studies, then these cohorts might also facilitate enrollment of high-risk individuals to clinical trials of medicines in the cancer prevention setting.

One route to the development of medicines for cancer prevention is to repurpose or reformulate existing medicines. For example, this approach was taken with endocrine therapy for the prevention of breast cancer (30). Alternatively, new agents for cancer prevention can be developed, with HPV vaccines being a good example (63). Developing new medicines has certain advantages, such as being able to drug new targets and to dial in other desirable properties, such as an improved safety profile or targeted delivery to the tissue of interest. In this subsection, we discuss several concepts relevant to pharmacological intervention to prevent cancer, including target selection and precision medicine, patient safety, treatment regimen (dose and schedule), drug delivery, and duration of therapy.

Target selection should be based on a strong scientific rationale that the drug will be active and safe in the cancer prevention setting. Supporting evidence that should be prerequisite before testing in human subjects includes (i) data demonstrating the importance of the target in the disease, (ii) preclinical and/or clinical evidence that the treatment could have activity, and (iii) appropriate preclinical and/or clinical safety data (265). The selection of patients based on the presence of a predictive biomarker (precision medicine) has been hugely impactful in developing effective treatments for malignant disease (266), and we would argue strongly for utilizing predictive biomarkers for cancer prevention trials as well whenever possible. This may be challenging and is highly dependent on the availability of both an established predictive biomarker for the drug and appropriate access to samples for biomarker assessment. One exciting area for precision medicine is agents to prevent cancer in cancer syndromes, where the targetable molecular drivers are generally known and can be detected via a simple genetic test (124). Two potential opportunities for precision cancer prevention are PARP inhibitors, which might be tested to prevent cancer in gBRCA1m and gBRCA2m carriers (129), or HIF2α inhibition, which might be tested to prevent cancer in VHL syndrome (130).

The safety and tolerability of medicines for cancer prevention is a significant challenge. Most individuals eligible for cancer prevention will be healthy and asymptomatic, and the use of medicines with serious side effects will be difficult to justify in terms of the risk–benefit ratio. Even if the drug is efficacious in the treatment of a premalignant lesion, unacceptable toxicity will be a barrier to drug approval, as the example of retinoids illustrates (22, 88). Even modest toxicities that are prolonged or that significantly impact QoL (e.g., alopecia, skin rash, nausea, or diarrhea) can pose a significant challenge to medication uptake, adherence, or compliance. One of the concerns that is a barrier to uptake of endocrine therapies, which are approved for cancer prevention based on their efficacy and safety profile, is a concern regarding side effects among eligible women (38). Furthermore, long-term monitoring of safety data is required beyond drug approval (4), as it may reveal adverse effects that were not initially evident, as was the case with COX-2 inhibitors (4, 106–109).

Therefore, the safety and tolerability profile of agents selected for cancer prevention must be exceptionally favorable. Significant consideration should be given to any known (or suspected) safety or tolerability issues before proceeding. Several strategies might be used to optimize the safety and tolerability of medication in the cancer prevention setting. One strategy involves exclusion of patients from the trial for whom there is a high suspicion that toxicity will be experienced, either due to prior knowledge (based on pharmacogenomics or a patient's medical history) or after a run-in period to identify patients who may be intolerant. As described below, other potential strategies include optimization of the treatment regimen or the use of innovative drug delivery technologies.

To optimize the safety and tolerability of medicines for cancer prevention, the medication might be administered (i) at a lower dose than the standard dose, (ii) on an intermittent rather than continuous schedule, or (iii) for a shorter time period. This is predicated on the drug still achieving efficacy when administered on the modified regimen. There are precedents for success in this regard: low-dose aspirin for the prevention of colorectal cancer (101), intermittent letrozole for the adjuvant treatment of breast cancer (267), and short-course immunotherapy in the preoperative setting (135, 136). When considering what type of alternative treatment regimen to pursue in a clinical trial setting, preclinical studies designed to test the efficacy and safety of alternative dosing regimens may be informative (129, 268, 269). In addition, the FDA recently announced Project Optimus, which is aimed at improving dose optimization in oncology drug development to minimize toxicity and maximize benefit to patients, and a recent article outlined several recommendations (270). These included utilizing pharmacodynamic biomarkers to support dose optimization, using randomized dose trials to evaluate multiple doses, and utilizing information beyond dose-limiting toxicities (including dose modifications and low-grade persistent toxicities that might affect tolerability; ref. 270). These same considerations may also be relevant to selecting a dose and schedule for cancer prevention.

It may be attractive to use innovative approaches to drug delivery in the cancer prevention setting, including locoregional, tumor-targeted, or controlled-release drug delivery. Theoretically, these approaches could be utilized to (i) maximize the efficient localized delivery of drug to the target cell population, (ii) reduce the chances of systemic toxicity, and/or (iii) improve adherence or compliance to treatment.

Locoregional drug delivery is where, instead of systemic administration, the medicine is physically delivered proximal to the tissue of interest, which may involve a specialized formulation or device (or a combination of both; refs. 271, 272). Importantly, there are examples of locoregional drug delivery that are already licensed for the treatment of premalignant disease: BCG for the treatment of bladder CIS is delivered directly to the bladder by intravesical administration (49), while agents for actinic keratosis are formulated as a cream, gel, or ointment for topical application to the skin (60). Another example of locoregional delivery that has been tested clinically, but that did not achieve approval, is inhaled corticosteroids for the treatment of bronchial dysplasia or lung nodules (273–275).

In cases where the lesion is localized to one organ, locoregional drug delivery theoretically has the potential to maximize delivery of the drug to the lesion while limiting systemic toxicity. For example, Goldberg and colleagues have described PRV111, a cisplatin-loaded polymeric matrix that when placed topically in the mouth releases cisplatin locally for the treatment of oral cancer (276). Preclinical studies demonstrated that PRV111 induced rapid and sustained regression of oral premalignant lesions, and that PRV111 produced significantly increased drug retention in tumors and lower systemic toxicity when compared with intravenous administration of cisplatin (276).

Tumor-targeted drug delivery refers to medicines administered systemically that are designed to localize with high specificity and selectivity to the target cells, such as antibody–drug conjugates (ADC; ref. 277) or nanomedicines (278). An ADC is a monoclonal antibody linked to a cytotoxic drug payload that has been designed to improve the therapeutic window by limiting cytotoxic drug exposure to only those cells that express the target antigen of the antibody (277). As of December 2021, 12 different ADCs were approved by the FDA for hematologic or solid tumor indications, mostly in locally advanced, metastatic, relapsed, or recurrent disease (279). Testing of ADCs in earlier disease settings, including for the treatment of premalignant lesions, might conceivably occur in the future. However, this will depend on being able to identify biologically compelling targets that are expressed selectively by premalignant lesions. In addition, the safety profile of ADCs may need to be improved substantially before testing in cancer prevention.

Controlled-release drug delivery involves a formulation or device that sustains drug release for a specified period of time at a predetermined rate (280). One prominent example is transdermal patches, which are used in a variety of nononcology applications, including both hormone replacement therapy (HRT) and nicotine replacement therapy (281). In theory, controlled-release systems may have several advantages compared with conventional drug administration routes (such as intravenous delivery or oral delivery using immediate-release tablets). Potential advantages of controlled-release may include a superior safety profile or improved adherence or compliance to medication. For example, a recent systematic review concluded that the risk of venous thromboembolism was reduced with transdermal patches compared with oral delivery in women receiving HRT (282). Other studies, which examined antihypertensive or contraceptive medication, found that adherence or compliance to medication was improved for transdermal patches compared with oral medication (283, 284). These findings suggest that controlled-release drug delivery might also be attractive in the cancer prevention setting, either to reduce toxicity or to improve adherence or compliance to medication. Controlled-release devices that can deliver agents intermittently might also be considered, given that preclinical studies to date indicate that intermittent dosing may be theoretically attractive for improving the therapeutic index of agents developed for cancer prevention (129, 268, 269).

To further illustrate the principle regarding how innovative approaches to drug delivery could be utilized for cancer prevention, in Fig. 3A–C, we provide three examples of how locoregional, tumor-targeted, or controlled-release drug delivery approaches might be utilized, theoretically, to deliver a medicine for the treatment of bronchial dysplasia. For further examples of drug delivery technologies, see Supplementary Table S3.

Duration of therapy may be dependent on several factors, including mechanism of action, disease biology, and clinical endpoint. For example, in terms of mechanism of action, while a vaccine might require only three shots to be effective, a drug inhibiting an oncogene (e.g., a tyrosine kinase inhibitor) may require more prolonged administration (weeks or months). In terms of disease biology, while a premalignant lesion might require only short-term therapy with a medicine to be eradicated, in other situations, such as cancer syndromes, a longer duration of therapy may be required. In terms of the clinical endpoint, if the primary goal is to regress a measurable premalignant lesion, then shorter duration of therapy might be sufficient. However, if the primary goal is to prevent the transition of premalignant lesions to malignant disease, then a longer duration of therapy may be necessary. It is important to consider the patient safety and tolerability issues that may arise with long-term administration of a drug. This is an area in which, potentially, the above-described considerations regarding dose and schedule, as well as drug delivery technologies, could provide some mitigation.

Phase I, II, and III studies can be utilized in cancer prevention in a similar fashion to how clinical trials are used in the development of medicines for advanced-stage disease (265). In the previous themes, we discussed the target population and the nature of the intervention. Below we discuss some specific further considerations for clinical trial design and execution.

The incidence of cancer (or cancer-free survival) was the primary outcome measure in many RCTs for cancer prevention, including pivotal studies of endocrine therapy for breast cancer prevention (31, 32). This typically involved running trials that were large, lengthy, and costly. However, there are several examples of “surrogate” or “early” clinical endpoints that can be measured within a shorter time frame (15, 285, 286). Surrogate endpoints can be histology-, imaging-, or liquid biopsy–based methods to evaluate whether the medicine is having the desired effect on the disease. They are attractive because they may be used during early-phase clinical development (phase I/II), including window-of-opportunity studies, to evaluate the activity of medicines quickly before then deciding whether to progress these same treatments to larger RCTs. Surrogate endpoints might even be considered as the primary endpoint in pivotal RCTs. In support of this, endpoints based on measuring the response of a premalignant lesion were the primary outcome measure in several pivotal studies that led to approvals of PDT/PS for BE with HGD (46) and topical medicines for actinic keratosis (60). However, this form of endpoint can also be contentious. Celecoxib received accelerated approval by the FDA on the basis that it reduced polyp burden. This was possible because polyp burden was considered a suitable and well-validated surrogate endpoint of clinical benefit (15). However, others have argued that demonstrating control of polyps alone may not be sufficient for regulatory approval (28).

There are other endpoints to consider in the cancer prevention setting. For example, testing whether a medicine can help patients avoid surgery may be clinically meaningful if it avoids patients having to undergo procedures that are associated with significant morbidity, such as cystectomy or radical prostatectomy. Another related endpoint could be based on QoL improvements, measured using patient-reported outcomes. An open question is whether trials in the cancer prevention setting should be designed to evaluate benefit in terms of overall survival. In most settings, this would involve designing trials that are prohibitively too large (in terms of size, duration, and cost) to make them viable in a drug development context. This further emphasizes the importance of identifying clinically meaningful endpoints that can be measured in a shorter time frame.

Selection of the primary endpoint requires careful consideration; it must be tailored to the specific disease setting and will require agreement with key stakeholders, in advance, that the endpoint is clinically meaningful. This includes discussion with regulators and payers to understand whether the endpoint will be valid for both registrational intent and reimbursement.

RCTs involving randomization to a medicine versus the standard of care will probably be necessary for most studies in cancer prevention. However, for several premalignant disease settings, the standard of care may vary, including regional differences. For example, while 49.9% of BRCA1/2 mutation carriers had prophylactic mastectomy to prevent breast cancer in the United States, in Poland it was only 4.5% of women (287). This heterogeneity means that careful oversight of what constitutes the standard of care, and careful site selection, may be necessary. In addition, patients may be motivated to use additional interventions to improve their long-term health and reduce their chances of developing cancer. This might include taking dietary supplements or electing to have risk-reducing surgery. The potential for these additional interventions to affect the integrity of either the comparator arm or the treatment arm, as well as strategies to mitigate the impacts, should be considered.

Recruitment to cancer prevention studies has challenges, and several publications have examined the opportunities and barriers for recruitment (288–293). From this work, several recommendations for optimizing recruitment may be synthesized. First, strategies should be used to find and screen as many potential participants as possible. This can include running studies at multiple sites, working with academic cooperative groups, and using digital tools or the media to reach potential participants. Second, strategies should be used to maximize the number that is eligible to participate. Eligibility criteria should not be overly restrictive and, wherever possible, should be designed to optimize recruitment regardless of age, race, and ethnicity. Third, potential participants should be provided with appropriate information regarding the benefits and risks of taking part in the study (both for themselves and wider societal impacts). Fourth, trial participation should be made as simple and convenient as possible for the participant. Fifth, recruitment should be actively monitored, so that the protocol can be modified to boost recruitment if necessary.

There are many reasons why individuals may not adhere to a prescribed medication, and several strategies to address the problem can be proposed (294). In the case of cancer prevention studies, the following could be helpful. The health care provider relationship with the patient should remain strong and supportive, including clear advice on the value of taking the medicine and pragmatic guidance on managing side effects, throughout the duration of the study. A variety of tools can also be used to remind participants to take their medication even though the individual may be feeling well. Published evidence shows that providing such support can promote adherence or compliance (295–297). In addition to more traditional methods, digital health solutions (298) could also be utilized to support patients.

As discussed below, the successful development of medicines for cancer prevention could be facilitated by changes in health care policy that (i) provide incentives for stakeholders, (ii) allow different pharmacoeconomic models to be utilized, and (iii) promote the importance of early detection and early treatment.

There are currently several disincentives for stakeholders to develop medicines for cancer prevention (22, 25). First, most registrational intent studies are typically large and costly studies involving thousands of patients, with lengthy follow-up required to meet regulatory endpoints (typically 5–10 years), which means that these trials are expensive to run. Second, once approval is finally obtained, the medicine may be nearing the end of its patent life, shortening the period of exclusivity that remains to recoup the initial investment. Third, because these trials involve asking healthy individuals to take a medicine, the potential for public liability claims is greater than a medicine developed for patients with advanced-stage cancer. Fourth, stakeholders may be hesitant to enter the field of cancer prevention because the regulatory path to approval for these agents is not straightforward.

In 2008, Grabowski and Moe recommended several policy reforms that could be implemented to address some of these issues (25). The three most salient recommendations were (i) to implement strategies that reduce the costs of research and development for cancer prevention drug development, including research subsidies or tax credits; (ii) to offer patent extensions to lengthen the exclusivity period for cancer prevention drug development (similar to the Orphan Drug Act); and (iii) to provide a no-fault liability insurance program for investigation of new drugs developed for cancer prevention (similar to the National Vaccine Injury Compensation Program; ref. 25).

Another potential strategy to address the patent issue, which may reduce the time from patent filing to market approval, would be to encourage stakeholders to move the most promising drug candidates into the cancer prevention setting earlier in the drug's clinical development (rather than waiting to do this at a later stage of development). A significant rate-limiting step will be having enough long-term safety data for the new molecule to support testing it in “healthy” individuals. In some instances, this might be circumvented by utilizing “next-generation” drug candidates, where there is already an established efficacy and safety database for the first generation of that drug class.

With regard to the uncertain regulatory path for cancer prevention, this concern might be resolved if regulatory authorities published further guidance for drug development in specific cancer prevention settings. For example, the FDA has published guidance to assist sponsors in the development of drugs for BCG-unresponsive NMIBC (299). We believe that similar documents with guidance for drug development in specific premalignant indications or high-risk individuals could incentivize the development of drugs for cancer prevention.

There will be challenges to face when seeking registration and reimbursement for drugs developed for cancer prevention. Conventional pharmacoeconomic models utilized by health care authorities, regulators, and payers are typically based on the treatment of advanced cancer. However, the proposition for cancer prevention is different, and therefore different pharmacoeconomic models will be needed to evaluate cost-effectiveness and clinical benefit. Models could be reinterpreted focusing on several factors, including (i) accepting the value of surrogate, early, or alternative endpoints; (ii) valuing the potential benefits for patients, mainly in terms of QoL improvements; and (iii) recognizing the potential benefits for the health care system and economy in reducing the cost of treating advanced-stage disease and the cost of end-of-life care. Further to this, the industry may need to take a flexible approach to the pricing of drugs for cancer prevention. This is not dissimilar to the approaches taken already to develop medicines in other therapeutic areas in which prevention has proven hugely impactful, such as the prevention of CVD with antihypertensives and lipid-lowering medication.

Medicines for cancer prevention will be impactful only in a health care environment that embraces the importance of early detection and early treatment and in which there is equity of access to these measures. A recently published Cancer Research UK road map document highlights several policy reforms that could promote success in this area, including an increased emphasis on early detection research, routine implementation of approved early detection technologies in health care systems, and moving toward a community-based health check system whereby disease can be detected early in healthy asymptomatic individuals (201). In addition, education for both physicians and the public regarding the importance of cancer early detection and early treatment will be important (300).

Although the development of medicines to prevent cancer is an important opportunity, it is also not without significant challenges. Here we have highlighted some general concepts that could enable or accelerate the opportunity under six key themes (also summarized in Fig. 4). We believe that these concepts both synergize with and build on previously published thinking in this area (4, 8, 9, 12, 22, 24, 25, 27, 28, 265). To implement these concepts effectively, alignment and collaboration between all relevant stakeholders (including academia, health care providers, industry, patient advocates, policy makers, regulatory authorities, and health care payers) will be exceptionally important.

All of the authors are, or were, employees of AstraZeneca at the time this work was performed and may have stock ownership and/or stock options or interests in the company. In addition, M. Moschetta reports other support from AstraZeneca and Novartis outside the submitted work, and S. Galbraith reports other support from AstraZeneca outside the submitted work.

We wish to acknowledge the late José Baselga for providing us with the initial inspiration to write this article. We also thank all of the scientists who provided input during the conceptualization or writing of the manuscript. In particular, we thank Anna Butturini, Phil Dennis, Tim Eisen, Guilia Fabbri, Matthew Farmer, Robert Godin, Matthew Hellmann, Elizabeth Kuczynski, Mark Landis, Bienvenu Loembe, Jacques Mascaro, Shethah Morgan, Andrew Mortlock, Mike Neidbala, Richard Olsson, Cecilia Orbegoso, Sarah Payne, Liz Pease, Kris Sachsenmeier, Tariq Sethi, Elizabeth Shaheen, Simon Smith, John Stone, Jill Walker, and Andy Williams. We thank Laura Prickett for assistance with figure preparation. We also thank the anonymous reviewers for their constructive feedback on the manuscript. Some parts of Figs. 1, 2, 3, and 4 were created with BioRender.com.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.