Sleep Patterns and Risk of Prostate Cancer: A Population-Based Case Control Study in France (EPICAP)

Prostate cancer is the second most frequent estimated cancer worldwide with more than 1.4 million new cases in 2020, and the first in industrialized countries among men (1). Among these countries, France is one of the country with the highest incidence rate. Today, we estimate that 66,000 new cases will be diagnosed each year with 9,000 associated deaths in France (2). Prostate cancer has become a major public health issue, but its etiology remains largely unknown, apart from some well-established risk factors such as age, ethnicity, and family history of prostate cancer (3).

The increasing prevalence of sleep disorders reported in industrialized countries since the last decades makes this phenomenon a major public health problem (4, 5). Insufficient sleep, poor quality of sleep, and the resulting circadian disruption could lead to many pathologies such as diabetes, depression, obesity, cardiovascular diseases, and even cancer (6). Another major source of circadian disruption, night shift work, was recently classified as “probably carcinogenic to humans” (Group 2A), on the basis of limited evidence of cancer in humans, including prostate cancer, sufficient evidence of cancer in experimental animals, and strong mechanistic evidence in experimental animals (7). Sleep is of crucial importance for the proper functioning of our body. From learning to memorization to the effectiveness of the immune system, sleep is involved at all levels of our well-being. The role of sleep deprivation has been associated in several chronic disease risk such as cardiovascular, metabolic diseases, or cancer (8). About 27% of the French population declared a sleep debt greater than 1 hour, 13% suffer from insomnia, and it is widely recognized that these sleep disorders can vary considerably over the life course (9).

The relationship between sleep and cancer risk is complex and epidemiologic studies showed inconstant results. Some studies have suggested an increased prostate cancer risk with shorter sleep duration or sleep disturbances (10–12), whereas others did not find any association (13–16). Recent studies also suggested that inadequate sleep duration, short or long duration, or sleep disruption could increase breast cancer risk in women (17–19), lung cancer risk (20, 21), or colorectal cancer (22). The exposure assessments used in these studies are not always comparable among them, and studies did not always take into account cancer aggressiveness or other circadian disturbance parameters, such as chronotype or night shift work, which further complicates the conclusions that can be drawn. Biological mechanisms underlying the potential association between sleep disorders and incidence of cancer include: a dysregulation of the production of melatonin which is known to inhibit cancer cell proliferation (23), a decrease of immune system and the activation of cancer-stimulatory cytokines (24), a systemic inflammation with increases in levels of C-reactive protein (CRP) and IL6 (25) or disruption and promotion of the activity of neuroendocrine stress systems (26).

It has been suggested that it would be relevant to have a better characterization of the different aspects of the circadian disruption, including various sleep patterns but also, chronotype and night shift work (27–29). Indeed, night shift work has been associated to prostate cancer risk in several studies (30–34), including one by our group (30), especially when chronotype was taken into account (15, 30, 31).

In this context, the aim of this paper was to investigate the role of sleep patterns in prostate cancer risk, taking into account circadian disruption and inflammation indicators, based on data from the EPICAP study.

EPICAP is a population-based case–control study conducted in the departement of Hérault in France. Details of this study have been published elsewhere (35).

Eligible cases were all men newly diagnosed for prostate cancer in 2012 to 2013, ages under 75 at the time of diagnosis, who resided in the departement of Hérault. Cases were recruited by trained clinical research nurses in all public hospitals and private urology clinics of the departement. Cancer cases included in the study were all histologically confirmed. Controls were men, randomly selected from the general population of Hérault, without a previous history of prostate cancer, and frequency-matched to the cases by 5-year age group at the time of recruitment. Quotas by age were applied to obtain the same distribution by age in the control group as in the case group (age frequency-matching). Quotas by socioeconomic status (SES) were applied to obtain a control group representative of the male population base with regard to SES (conditionally to age), and were used to minimize selection biases that may arise from differential participation rates across SES categories.

Overall, we recruited 819 incident prostate cancer cases and 879 population controls with a participation rate of 75% and 79%, respectively.

The EPICAP study was approved by the ethical committee of the French NIH and medical research (INSERM, international review board no. 01–040, November 2010) registered with the Office for Human Research Protection (OHRP) of the United States Department of Health and authorized by the French data protection authority (CNIL no. 910485, April 2011). All participants included in the study provided a written informed consent.

Data pertaining to each study subject were obtained from a face-to-face standardized computerized questionnaire (Computer Assisted Personal Interview, CAPI) realized by research clinical nurses. We collected information on sociodemographic characteristics, family history of cancer, residential and occupational history, personal weight and medical history, lifetime consumption of tobacco and alcohol, and recreational activities such as physical activity. Tobacco consumption was classified into three categories: never, former, and current smokers. For alcohol consumption, men were asked during interview whether they drink more than once a month during 1 year (no/yes). For those who answered “Yes,” the level of alcohol consumption was assessed using the CAGE questionnaire (36). Alcohol consumption was classified into three categories: never drinkers (less than once a month during 1 year), low drinkers (at least once a month during 1 year and zero or one positive answer to the CAGE questionnaire) and heavy drinkers (at least once a month during 1 year and two or more positive answers to the CAGE questionnaire). Physical activity was assessed using metabolic equivalent task (MET, hour/week/year; ref. 37) for each activity that has been practiced at least 1 hour per week during 1 year. We classified this variable into quartiles, corresponding to physical intensity over the life course, calculated in the control population: Q1 (<6.25 MET-hour/week/year), Q2 (6.25–13.0 MET-hour/week/year), Q3 (13.0–24.15 MET-hour/week/year), Q4 (≥24.15 MET-hour/week/year).

Among cases, Gleason scores, prostate specific antigen (PSA) levels, and tumor stage at diagnosis were extracted from patient's medical records and validated by the Hérault Cancer Registry (38).

Data related to sleep patterns were retrospectively collected since early adulthood. Subject were asked “How many hours per night on average did you sleep during the week?” before age 30, between age 30 and 50, between age 50 and 60, and since age 60. The same question was also asked for duration of sleep during week end (considered as a proxy of sleep duration during days off). Then, we calculated the average duration of sleep at each period of life [average duration during week and week end, weighted for the number of days in week (5) and the number of days in week end (2)]. The lifetime mean sleep duration was finally obtained by calculating the average sleep duration over the four periods. We divided duration of sleep into five groups (<6 hours, [6–7[ hours, [7–8[ hours, [8–9[ hours, ≥ 9 hours) and used “[7–8[ hours” of sleep duration for reference group. Sleep deprivation was defined by calculating the difference between the duration of sleep during weekend and week for each period and the lifetime mean sleep deprivation was obtained by calculating the average deprivation over the four periods. This definition has been widely used in the literature and few papers examined sleep deprivation and its association to cancer risk (39–41). Moreover, this indicator has recently been shown to be a good indicator of sleep disturbance compared with sleep duration (42). Sleep deprivation was classified into three groups (no deprivation or <1 hour as reference group, 1–2 hour, ≥2 hours) or two groups for some analysis where numbers were too small (no deprivation, ≥1 hour). To collect data on sleep medication use, subjects were asked if they had ever taken any pills for insomnia disorders, prescribed by a physician (sleeping pills or anxiolytics) for a duration of at least 1 year (no or <1 year of sleep medication as reference group, ≥1 year of sleep medication). Duration of all medication use were also collected and allowed us to calculate the total duration of sleep medication use over lifetime. Duration of sleep medication use was categorized into four groups (reference group: no medication or <1 year of sleep medication, <10 years, 10–19 years, ≥20 years) or three groups when numbers were too low (no medication or <1 year, <10 years, ≥10 years). Chronotype was assessed using the Morningness-Eveningness Questionnaire (43) adapted to the French population, allowing us to classify each subject as a morning, evening or neutral (undifferentiated) person, according to the classification from Taillard and colleagues (44). Information on work time was obtained for each job held for more than 6 months. Men were considered as night worker if they had worked for at least 270 hours of night work per year or three nights per month during at least 1 year, according to the French legal definition (30). Duration of night shift work (total number of years in jobs involving night work) was assessed for night workers and categorized into three groups according to a cut off close to the median of duration of night shift in the control group (no night work, night work <15 years, night work ≥15 years). NSAID use was collected and classified as nonusers (men who never took, or less than once a month) or ever users (at least once a month; ref. 45).

ORs and 95% confidence intervals (95% CI) were calculated using unconditional logistic regression models. Polytomous logistic regression was also used where cases were subdivided according to prostate cancer aggressiveness (low or intermediate aggressiveness: Gleason score ≤3 + 4 vs. high aggressiveness: Gleason score ≥4 + 3).

All analyses were systematically adjusted for age at reference date, age at diagnosis for cases, and age at recruitment for controls (5 years age group), family history of prostate cancer in first-degree relatives (yes/no), ethnic origin (Caucasians, non-Caucasians). Analyses were also adjusted for potential confounding factors such as educational level (no diploma, less than high school, high school graduate, college graduate), body mass index measured during the interview (BMI, <25 kg/m², 25–29.9 kg/m², ≥30 kg/m²), night shift work (never/ever), and chronotype (morning, neutral, evening type).

On the basis of possible biological mechanisms that may explain the associations between sleep patterns and prostate cancer risk and based on previous results regarding night shift work, chronotype (30, 46), and NSAIDs use (45, 47) in the EPICAP study, we decided to stratify the analyses by chronotype (morning, neutral, evening type), night shift work (no night work, night work <15 years, night work ≥15 years) and NSAIDs use (never/ever). P values for interaction were tested using the Wald test.

All statistical analyses were performed using SAS (version 9.4; SAS Institute Inc.).

The datasets analyzed during this study are publicly available from the corresponding author on reasonable request.

The description of cases and controls by sociodemographic characteristics and suspected prostate cancer risk factors are shown in Table 1. As expected, prostate cancer risk was highly associated with family history of prostate cancer in first-degree relatives (P < 0.0001). Cases and controls did not differ with respect to age, ethnic origin, educational level, tobacco or alcohol consumption, BMI, physical activity, night work, or chronotype.

Table 2 shows the determinants of mean sleep duration (<7 hours / ≥7 hours) in the control group. Sleep duration below 7 hours per night was positively with associated higher BMI (P = 0.005), lower educational level (P = 0.007), night shift work (P = 0.001), and morning chronotype (P = 0.003). No difference was observed for tobacco or alcohol consumption, physical activity, or family history of prostate cancer. No association was seen between sleep duration and sleep deprivation (P = 0.114) or medication use for sleep (P = 0.234).

Table 3 shows associations between sleep indicators and prostate cancer risk, overall and according to prostate cancer aggressiveness. Mean sleep duration and mean sleep deprivation were not associated with prostate cancer risk, either for overall prostate cancer or for less aggressive prostate cancer. A U-shape curve was observed when we analyzed the association between sleep duration and risk of aggressive prostate cancer, but ORs were not significant. Similar results were observed considering sleep duration by age periods (<30, 30–50, 50–60, >60 years, and current). Sleep medication use was associated with prostate cancer risk with a nonsignificant OR of 1.20 (95% CI, 0.90–1.62) and the OR reached 2.11 (95% CI, 0.92–4.83) for duration of use equal or longer than 20 years (P_trend for continuous variable = 0.0085).

Associations between prostate cancer and sleep indicators after stratification by chronotype are presented in Table 4. We observed a positive association for very short duration of sleep among evening chronotype (OR, 4.88; 95% CI, 1.33–17.9 for mean sleep duration <6 hours) with a nonsignificant interaction between duration of sleep and chronotype. Among these men, we also observed an increased risk of prostate cancer for sleep deprivation ≥1 hour (OR, 1.96; 95% CI, 1.04–3.70). P value for interaction between the sleep deprivation and chronotype on prostate cancer risk was 0.014. Association between sleep medication use and prostate cancer risk was restricted to neutral type (OR, 1.52; 95% CI, 1.00–2.29 for ever sleep medication use). ORs increased for longer duration of sleep medication use for all chronotype without reaching significance. Nevertheless, P_trend for duration of sleep medication use was close to significance for the group of neutral chronotype (P = 0.07).

Associations between sleep indicators and prostate cancer risk, stratified on night shift work duration are shown in Table 5. Among men who ever worked at night, increased ORs were observed for duration of sleep ≥9 hours only (OR, 3.74; 95% CI, 0.73–19.2 for work night work <15 years; OR = 4.29; 95% CI, 0.80–22.9 for night work ≥15 years) even though not significant due to small numbers. When combining the two categories of night workers, OR reached 3.91 (95% CI, 1.28–12.0). We observed an association between prostate cancer risk and duration of sleep medication use ≥10 years (3.84; 95% CI: 1.30–11.4; P_trend = 0.0188). We did not observe any association for men who never worked at night.

Table 6 shows associations between sleep indicators and prostate cancer risk, stratified on NSAID use. No difference was observed for sleep duration or sleep deprivation between the two strata. We observed an association between prostate cancer risk and sleep medication among nonusers of NSAID (1.45; 95% CI, 0.98–2.15; P_interaction = 0.08), and an association for long use of sleep medication (OR for use ≥10 years 2.08; 95% CI, 1.15–3.75), with a P_trend for duration of sleep medication use highly significant (P = 0.0037).

We performed a sensitivity analysis restricted to controls who underwent PSA-based screening in the last 2 years, the results were unchanged (Supplementary Table S1).

Our results are based on a population-based case–control study specifically designed to address circadian disruption and prostate cancer risk. We did not find any association between sleep duration or sleep deprivation and prostate cancer risk. However, we observed a significant increased risk of prostate cancer with the increase of duration of sleep medication. In addition, our results suggest that chronotype, night shift work, or NSAID could modify the relationship between sleep duration, sleep deprivation, or sleep medication use and risk of prostate cancer.

Results from epidemiologic studies published to date have been inconsistent and still rare for prostate cancer. Few previous studies have found an increased risk of prostate cancer for short duration of sleep (10, 48), whereas others did not find any association (13–16). However, to our knowledge, we are the first study to examine sleep deprivation as a sleep pattern indicator. Other cancer sites, mainly breast cancer, have been more extensively studied and have shown interesting results. In women, some studies suggested that long sleep duration could be associated with a decreased risk of breast cancer (17, 39, 49–52), however a recent meta-analysis showed an increased risk of breast cancer with longer duration (53). Besides, one Chinese study showed increased risks of breast cancer for both short and long duration of sleep (19). To date, very few studies analyzed the effect of sleep deprivation on cancer risk. To our knowledge, only two studies on breast cancer described a modest and nonsignificant association with sleep deprivation (39, 41). Our results did not show any significant relationship between sleep deprivation and overall prostate cancer risk. When restricted to evening chronotype, men with sleep deprivation greater than 1 hour presented a two-fold risk of prostate cancer.

Interestingly, the U-shape curve observed for duration of sleep has also been observed in two studies on lung cancer, where short and long duration of sleep were associated with increased cancer risk (20) and mortality (21). In women, Jiao and colleagues showed this U-shaped relationship between sleep duration and colorectal cancer risk (22). In the same way, it has been observed that both short and long duration of sleep were associated with all-cause mortality (6). In 2016, results of the large American Association Retired Person cohort from the NIH also suggested associations with both short and long sleep duration for many cancer risks in both genders (54). In our study, it can be noted that this curve shape emerge when we analyzed the association between sleep duration and risk of aggressive prostate cancer, even though not significant. Moreover, a positive association was observed for long duration of sleep when analyses were restricted to night workers. Such a result was observed in an epidemiologic study, which showed a synergistic effect of longer duration of sleep and night work on breast cancer risk (19).

The biologic mechanisms underlying the potential association between sleep duration and incidence of prostate cancer is complex and not fully understood. Short sleep or sleep deprivation are linked to a dysregulation of the production of melatonin, which is known to inhibit prostate cancer cell proliferation (23, 55). It has been showed that men with sleep problems had lower morning levels of urinary 6-sulfatoxymelatonin, a major metabolite of melatonin, which has been associated with increased risk of advanced prostate cancer (56). Another hypothesis is that sleep deprivation can also lead to immune suppression, activate cancer-stimulatory cytokines, and finally promotes cancer (24, 57). In contrast, long sleep duration could be a marker of ill-health, rather than a direct effect in cancer risk (6). Indeed, depressive symptoms, low socioeconomic status, higher BMI, or poor general health have been shown to be related with long duration of sleep (58–60).

Sleep quality is reported in various ways according to studies, from a simple question “Do you sleep well?” to a precise diagnosis of sleep disorder from health administrative database. In our study, sleep medication was used as a surrogate of sleep disturbance. Some studies have found increased risk of prostate cancer for sleep disorders (11, 12) and some suggested that hypnotics use may be relate to risk of prostate cancer (61, 62). Although others have found no association with sleep disturbance (10, 14–16, 63). Interestingly, a recent meta-analysis showed that sleep disturbance could be associated with increases in markers of systemic inflammation such as CRP or IL6 (25). In our study, we found an elevated risk of prostate cancer with the increase of duration of sleep medication and this association increased for men who never use NSAID, which supports the thesis of the inflammatory effect of sleep disorders.

In this population-based study, case identification was done in all private and public cancer hospitals in the the département of Hérault. The number of eligible cases identified was close to the expected number based on age-specific incidence rates for France (38). Even though participation rate in cases was 75%, the age distribution and the Gleason score were comparable with those of the Hérault Cancer Registry. The controls were randomly selected from the general population of the département and we used quotas defined for age and SES for controlling possible selection bias.

Epidemiologic surveys indicate that 11% to 17% of people over 65 years of age used sleeping pills, with or without a prescription (64–66). These numbers are quite similar to the 13.6% of sleep medication users in our control population above 65 years. Our control population also showed a mean sleep duration of 7 hours 10 minutes, equal to the results of the French 2017 Health Barometer Survey showing that the mean of total sleep time (without taking into account nocturnal awakenings) was estimated to 7 hours 12 minutes in the male population (9). Our data are therefore quite consistent with those found in French literature, and classification bias seems unlikely.

As in any case–control study, recall bias could not be ruled out. However, this was reduced in our study by the use of standardized questionnaires by trained interviewers, similar interview conditions for cases and controls, and the use of a precise questionnaire for eliciting information on sleep characteristics. Moreover, results on sleep duration by age periods were similar whatever the period, including more recent age periods (i.e., 50–60, >60, or current), which are less susceptible to error measurements. For men concerned by social jetlag or irregular working hours, we cannot entirely exclude that sleep duration on weekdays and weekends do not accurately reflect the sleep durations on work days and free days. However, the analysis has been performed on the average sleep duration and sleep deprivation within lifetime which minimize this issue. All sleep indicators that have been investigated in this study were self-reported and not objectively measured, which may have led to misclassifications. We cannot ruled out that the null results of the main analysis regarding sleep duration as well as sleep deprivation could be driven by such nondifferential measurement error biases, and on the other hand, this classification bias is not a likely explanation for the observed associations in the stratified analyses. Our study was of relatively large size, but the statistical power may have been limited in some stratified analyses. To minimize prostate cancer screening bias for controls, we restricted analysis for controls who have been screened for prostate cancer for less than 2 years. The estimates for long duration of sleep, mean sleep deprivation, or duration of sleep medication were slightly increased but confidence intervals were wider.

Some authors suggested that sleep deprivation or sleep quality could be a better indicator than duration of sleep to capture sleep disturbance (63). The major strengths of the present study are the various indicators of sleep disturbance over the lifetime, with 4 measurement points (before 30, between 30 and 50, between 50 and 60, and after 60 years old), duration of sleep during work days and days off, and detailed information on sleep medication use. We were also able to focus on factors associated with sleep and its disturbances such as night shift work over the entire lifetime or chronotype, defined by a validated scale. In addition, all potentially important confounders were taken into account in the analysis, and results remain unchanged after adjustments.

In conclusion, our study suggests an association between sleep disturbance and prostate cancer risk for some male profiles. We suggested that some factors such as chronotype, night shift work or NSAID use could modify the relationship between sleep disturbance and risk of prostate cancer. These findings point to the importance of examining several indicators of sleep disturbance in a global context of circadian disruption. Further studies exploring the impact of sleep disturbance on prostate cancer risk should consider all of these parameters to better understand their effects and interactions.

No author disclosures were reported.

E. Cordina-Duverger: Formal analysis, methodology, writing–original draft, writing–review and editing. S. Cénée: Formal analysis. B. Trétarre: Resources. X. Rebillard: Resources. P.J. Lamy: Resources. G. Wendeu-Foyet: Data curation. F. Menegaux: Conceptualization, data curation, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing.

The EPICAP study was funded by Ligue nationale contre le cancer (to F. Menegaux), Ligue contre le cancer du Val-de-Marne (to F. Menegaux), Fondation de France (to F. Menegaux), Agence nationale de securite sanitaire de l'alimentation, de l'environnement et du travail (ANSES; to F. Menegaux). We would like to thank the clinical research nurses who were in charge of participants’ interview, anthropometric measurements, and biological sample collection (Anne-Laure Astolfi, Coline Bernard, Oriane Boyer, Marie-Hélène De Campo, Sandrine Margaroli, Louise N'Diaye, Sabine Perrier-Bonnet). We also would like to thank Christian Prad and Nadine Soller for help with patient medical data collection within the Hérault Cancer Registry (Registre des tumeurs de l'Hérault, Montpellier, France). We are grateful to the EPICAP study Group: Urologists: Drs Didier Ayuso, Bruno Segui (Centre Hospitalier Bassin de Thau, Sète, France), Alain Guillaume, Jean-Paul Constans, Olivier Delbos, Pierre Lanfray, Damien Rizet, Etienne Cuénant (Cabinet Urologie du Polygone, Montpellier, France), Michel Locci (Centre Hospitalier, Béziers, France), Etienne Cuénant (Clinique Ste Thérèse, Sète, France), Nicolas Drianno, Bernard Marc, Paulo Soares (Polyclinique Saint Privat, Béziers, France), Antoine Faix, Samer Abdel Hamid, Bruno Segui (Service urologie, Clinique Beau Soleil, Montpellier, France), Samer Abdel Hamid (Clinique Saint Louis, Ganges, France), Laurent Cabaniols, Maxime Robert, Rodolphe Thuret (Centre Hospitalo-Universitaire, Hôpital Lapeyronie, Montpellier, France). Pathologists: Drs Didier Brel, Lysiane Schweizer, Philippe Nayraud, C. Lecam-Savin (Béziers), Roland Daniel, Jean Baptiste Perdigou, Chantal Compan, Mireille Granier, Jean Louis Bouzigues, Elisabeth Broquerie, Joëlle Simony, Frédéric Bibeau, Pierre Baldet, Isabelle Serre (Montpellier), Marie Laure Gaume (Sète).

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.