Influence of Primary Care Use on Population Delivery of Colorectal Cancer Screening

Although screening can reduce colorectal cancer (CRC) mortality (1), nearly half of eligible patients in the United States have not been screened (2). Screening can be accomplished via several testing options, including fecal occult blood testing, sigmoidoscopy, and colonoscopy (3), but completion of each test is a multistep process, typically initiated by a primary care physician's order or by a primary care physician's referral to a specialist. Thus, a patient-physician encounter, particularly within primary care, may be a crucial first step in the process of CRC screening, and patients who attend few or no primary care visits may be especially likely to remain unscreened.

Indeed, a primary care physician's recommendation and assistance are consistent and powerful predictors of patient completion of CRC screening (4-7). However, it is possible that a substantial fraction of the unscreened population receives little or no primary care, and so has little or no opportunity to receive a physician's recommendation. Understanding how the unscreened population is presently served by primary care can assist health plans in judging the relative importance of opportunistic and outreach strategies to increasing screening. If limited or absent primary care use is common and strongly associated with a lack of screening, then health systems might invest in efforts to promote primary care attendance. On the other hand, if most patients attend primary care visits yet still go unscreened, resources might be better devoted to maximizing the opportunistic delivery of CRC screening during primary care visits that patients are already attending.

We estimated the extent to which limited or absent primary care use contributes to underuse of CRC testing within the population served by a large U.S. health plan. Because the health plan collects relatively comprehensive utilization data for all enrollees, we could ascertain completion of CRC testing across an entire population, including patients with few or no primary care visits. We hypothesized that receipt of little or no primary care would constitute a substantial barrier to population delivery of CRC testing. In addition, we hypothesized that the contribution would be relatively greater among men at onset of screening eligibility (age 50 years) because, unlike women, men in their early fifties may not be accustomed to routine physician visits for preventive care.

Subjects were enrolled in Group Health, a prepaid health plan that serves ∼350,000 enrollees in western Washington State. Data sources to determine eligibility and study variables included automated health care and pharmacy data and a regional Surveillance Epidemiology and End Results cancer registry, which have been used extensively for research. The Group Health Human Subjects Review Committee approved the study methods.

We identified a population-based cohort of enrollees who were ages 50 to 78 years on January 1, 2002 and who were eligible for CRC screening in 2002 to 2003 based on absence of sigmoidoscopy, colonoscopy, or barium enema in 1997 to 2001. Subjects were continuously enrolled during the study years and five prior years to enable ascertainment of CRC testing from 1997 to 2003. Additionally, subjects lacked diagnostic indications for surveillance or diagnostic CRC testing [i.e., any positive fecal occult blood test (FOBT) in 1997-2001, a history of prior CRC, colonic polyps, or inflammatory bowel disease]. CRC diagnoses were identified by Surveillance Epidemiology and End Results registry linkage. Diagnostic and procedural codes used to identify diagnoses and CRC tests have been previously reported (8, 9).

Enrollees either select or are assigned primary care physicians and are encouraged to seek most initial care from primary care physicians located at 20 clinics across the Puget Sound region. Although clinic staff schedule new appointments with assigned primary care physicians whenever possible, patients may seek and receive care from other plan physicians (including specialists). In 2002 to 2003, plan recommendations regarding CRC screening for adults ages 50 y and older were consistent with the 2002 recommendations of the U.S. Preventive Services Task Force but emphasized regular FOBT and periodic sigmoidoscopy (10). Primary care physicians, however, could refer patients for screening colonoscopy.

We classified subjects as unscreened if none of the following 4 CRC tests were completed in 2002 to 2003: FOBT, sigmoidoscopy, colonoscopy, or double-contrast barium enema. FOBT completion was ascertained using automated laboratory data. Sigmoidoscopy, colonoscopy, and barium enema tests were identified from procedural codes on outpatient and inpatient encounters. Due to concerns about misclassification (11), we did not attempt to distinguish screening from diagnostic CRC tests.

Health plan databases define adult primary care providers based on generalist specialty (i.e., family medicine, general internal medicine, or generalist nurse practitioners or physician assistants) and practice location at one or more primary care clinics. We used automated health care data to count the number of visits to primary care providers and classified patients has having 0, 1, 2, 3, 4, 5, 6 to 7, or ≥8 visits in the 2-year study period. We categorized primary care use in this fashion to allow dose-response analyses by exposure to primary care. We selected an upper cutoff of ≥8 visits in 2002 to 2003 because only 40% of the population had ≥8 visits, so these patients are above the population median in primary care use. Encounters with obstetrician-gynecologists were not considered primary care visits because these clinicians do not function in this capacity in the plan.

We analyzed outpatient and inpatient encounter data to compute an automated form of the Charlson comorbidity index (12). We determined counts of nonprimary care visits (including specialist and emergency visits but not mental health visits). We also developed two variables to reflect preventive health behavior in the previous 2 years (2001-2002) because we believed these would correlate with unmeasured patient attitudes and beliefs regarding preventive health care. First, we counted the number of FOBTs completed in 2001 to 2002. Second, we determined whether patients received a preventive health examination in 2001 to 2002 as defined in previous research (8, 13). To develop proxy measures of socioeconomic status (i.e., median household income and proportion of persons ages 25 y and older that completed high school in the residence ZIP code; ref. 14), we linked subjects to a health plan database of year 2000 Census data. The large majority of subjects (95.1%) were successfully linked to Census data; no patient covariates were significantly associated with successful Census linkage (including age, sex, comorbidity, and health care use variables described above).

The analytic goal was to estimate the population attributable risk (PAR%) of going unscreened for CRC associated with decreasing exposure to primary care. PAR% estimates the proportion of disease that is attributable to a risk factor in a population and, therefore, provides policymakers with useful information about the value to population health of devoting resources to ameliorate or remove the risk factor (15). An important property of PAR% is that it depends not only on the strength of the risk factor, expressed as relative risk but also on the prevalence of the risk factor among diseased persons. In our context, the “disease” is lack of screening and the “risk factor” is a lower versus a higher number of primary care visits, and PAR% is calculated as:

where RR is the adjusted relative risk of going unscreened associated with a specific number of primary care visits (relative to the reference of ≥8 primary care visits) and P is the proportion of unscreened subjects with that number of primary care visits (16).

We first performed descriptive analyses to identify covariates that were significantly associated with limited primary care use. In particular, we used χ² tests to compare the distribution of covariates among patients with 0 or 1 primary visits and ≥2 primary care visits. We chose these visit categories because relative risks of lack of screening were found to be particularly elevated for those with 0 or 1 primary care visit.

Using random-effects logistic regression, we then estimated the relative odds of going unscreened associated with different numbers of primary care visits while adjusting for age (in 5-year categories), sex, Charlson comorbidity index (0, 1, 2, ≥3), nonprimary care visits (quintiles), and number of FOBTs (0, 1, ≥2) and preventive health examination receipt in 2001 to 2002. We used random-effects models to correct SEs for clustering of patients within primary care physician practices. We then used fitted models to predict adjusted percents of unscreened patients within each primary care visit stratum among both women and men (17). Using these adjusted percents, we computed the adjusted relative risk of lack of screening as the ratio of the predicted percent unscreened within each strata and the predicted percent of patients with ≥8 primary care visits that were unscreened (the reference).

We then estimated the adjusted proportion of all unscreened women and men that were within each primary care visit stratum (i.e., P in the formula for PAR% given above). First, we estimated the adjusted number of men or women in each primary care visit stratum as the product of the predicted percent unscreened (from our logistic model) and the total (crude) number of women and men within the stratum. We then estimated the adjusted proportion of the total number of unscreened women and men that were in the each visit stratum as the ratio of the adjusted number of women and men in each strata and the total number of women and men in all visit strata. Using the adjusted relative risks and adjusted proportions of the unscreened population within each primary care visit stratum, we finally computed the PAR% for each primary care visit stratum for both women and men.

Whereas initial analyses grouped primary care visits discretely (i.e., 0, 1, 2, 3, etc.), additional models grouped those with the lowest number of primary care visits (i.e., 0 to 1, 0 to 2, 0 to 3). These models allowed us to calculate the PAR% using more or less stringent definitions of limited primary care visits and to assess whether there was a threshold of primary care exposure beyond which the PAR% was very small. We also calculated the PAR% among women and men of different ages to judge whether limited primary care made a relatively greater contribution to lack of screening among younger men or women. Finally, among the subjects successfully linked to Census data (n = 46,324), we repeated the analyses while additionally adjusting for neighborhood median household income and educational status. Because results were essentially unchanged, we report results among the full sample without these adjustments.

Among 48,712 women and men who were eligible for CRC screening, the mean number of primary care visits in 2002 to 2003 was 8.7 (median, 6; SD, 9.7). Women had a greater mean number of primary care visits than men (9.6 versus 7.6 visits; pairwise comparison, P < 0.001), and a smaller proportion of women had fewer than 2 visits than men (10.3% of all women versus 17.0% of all men; P < 0.001). Compared with patients with two or more primary care visits, patients with zero or one visits were significantly more likely to be younger and of lesser comorbidity (Table 1). Additionally, patients with 0 or 1 primary care visits had fewer nonprimary care visits and were less likely to have completed FOBTs or attended preventive health examinations in 2001 to 2002.

Among both women and men, relative risks of going unscreened in 2002 to 2003 were highest among those with zero primary care visits and declined with an increasing number of primary care visits (Table 2). Indeed, the adjusted percents of women and men with zero visits that were unscreened exceeded 98%. Nevertheless, even among women and men with ≥8 primary care visits, over half of patients remained unscreened. Relative to patients with ≥8 visits, the risk of being unscreened was ∼60% greater among women and 70% greater among men with 0 primary care visits. However, a majority of unscreened women and men attended ≥4 primary care visits in 2002 to 2003, whereas a minority of unscreened women (14.3%) and men (23.1%) had 0 or one primary care visits. As a result, the PAR% of 0 visits was only 3.0% among women and 5.6% among men. Similarly, the PAR% of having 1 primary care visit was 1.9% among women and 3.6% among men.

Because a greater proportion of unscreened men had relatively few primary care visits than women, the PAR% associated with low visits was higher among men than women whether low visits was defined as 0, 0 to 1, 0 to 2, or 0 to 3 primary care visits (Table 3). For example, the PAR% associated with attendance of 0 to 3 visits was ∼70% greater among men than women (11.5% versus 6.8%, respectively). Meanwhile, among men, receipt of few visits was especially common among unscreened younger men. Thus, the PAR% associated with 0 to 3 visits among men ages 50 to 54 years was 12.1% compared with 5.4% among men ages 75 to 78 years. Among women, a similar fraction of younger and older unscreened women received few visits. Thus, in contrast to the male population, the PAR% associated with few visits was generally similar among younger and older women.

Among a population-based sample of health plan enrollees eligible for CRC screening, most patients who went unscreened over a 2-year period received several primary care visits during that time. Although receipt of few primary care visits was associated with lack of screening, the fraction of lack of screening that could potentially be attributed to limited primary care exposure (i.e., 0 to 3 visits over a 2-year period) exceeded 10% only among men. Among women, the fraction of lack of screening associated with limited primary was substantially less because a large majority of unscreened women had more than three primary care visits. In similar settings, policymakers and health system leaders may wish to prioritize interventions to optimize opportunistic delivery of CRC screening during existing primary care visits, rather than outreach efforts to encourage primary care attendance among population subgroups that are accessing little or no primary care. Alternatively, programs to promote CRC screening independently of primary care may hold promise.

Research suggests several potential targets for improving the delivery of CRC screening during primary care visits. Because patients consistently cite the importance of a physician's recommendation in motivating CRC screening (4-7), health systems should prioritize interventions with proven efficacy in increasing physicians' CRC screening recommendations, including the following: (a) educational programs to increase provider awareness of CRC screening guidelines (18); (b) systems to prompt patients to inquire about CRC screening (19-21) or to remind providers at the point-of-care regarding patient eligibility (20, 22, 23); (c) financial incentives for providers or patients (24, 25); and (d) promotion of visits dedicated to the delivery of evidence-based preventive services (8, 24, 26). Primary care systems could also augment CRC screening by assigning nonphysician team members (e.g., nurses or medical assistants) tasks of assessing and counseling patients regarding CRC screening eligibility (27-29). Finally, there is likely room for improving the effectiveness of provider counseling regarding CRC screening, as many CRC screening discussions apparently do not conclude with patient uptake of CRC screening (30, 31).

Although the strength of the association between limited primary care and absence of CRC screening was similar among women and men, the PAR% was greater among men than women because a greater proportion of unscreened men had few primary care visits. In addition, the PAR% of limited primary care was highest among younger men (ages 50-59 years). Whereas many women ages 50 to 59 years may be accustomed to seeking regular primary care services for breast and cervical cancer screening, many men of the same age may be unaware of guidelines recommending CRC screening for adults ages 50 years and older. In addition, younger men may perceive the attendance of a preventive physician visit as contrary to current social norms, suggesting the need for public health messages encouraging men over age 50 years to discuss CRC screening and other evidence-based preventive services with primary care physicians.

Relative risk estimates in our study may be subject to residual confounding by unmeasured patient characteristics that are associated with both primary care attendance and completion of CRC screening. To the extent that unmeasured patient factors explain observed associations rather than visit attendance, our study may exaggerate the true PAR% associated with low numbers of primary care visits. In addition, CRC tests may have been performed for diagnostic rather than screening purposes. Because patients with more primary care visits may be more likely to have tests performed for diagnostic purposes, we may have overestimated the true percent of patients who received CRC screening to a relatively greater extent in strata of patients with greater numbers of primary care visits. Such differential misclassification would tend to inflate PAR% estimates by increasing both the relative risk associated with low visits and P (the proportion of all unscreened patients that had low visits). In spite of potential biases that may inflate PAR% estimates, PAR% estimates were generally small, so the policy implications of our findings do not seem altered.

The comprehensiveness of the health plan data (including data on patients who received zero primary care visits) enabled us to quantify the effect of primary care use on the delivery of CRC screening to an entire population. Nevertheless, the population was insured by a single integrated health plan with a relatively high degree of primary care access. In other populations with lesser access to primary care, limited primary care use may contribute to a greater extent to slow population uptake of CRC screening.

Among health plan enrollees, limited primary care attendance was associated with a lack of CRC screening, yet most unscreened women and men attended regular primary care over a 2-year period. Our study therefore suggests that health plan efforts to increase population use of CRC screening might best prioritize the successful delivery of CRC screening during primary care visits that patients already attend.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.