Factors Associated With Low-Value Cancer Screenings in the Veterans Health Administration

Key Points Question What rates and factors are associated with low-value screenings (outside clinical guidelines) for common cancers in the Veterans Health Administration? Findings In this cohort study of 5 993 010 veterans, less than 3% of cancer screening recipients received a low-value test for breast, cervical, or colorectal cancer, but 39% of men screened for prostate cancer received a low-value test. No single factor (patient, clinician, clinical, or organizational characteristic) was associated with low-value screening receipt across these 4 cancer types. Meaning Findings suggest that low-value prostate cancer screenings are common in the Veterans Health Administration, although factors associated with low-value screenings differ by cancer type.


Introduction
Health care without benefit or in which the potential harm outweighs the benefit is considered low value. Cancer screening can become low value, for example, with increasing age, greater illness burden, or lower life expectancy 1,2 ; in these scenarios, short-term risks (such as procedural complications or testing burden) outweigh the expected benefits from detecting slow-growing cancers. [3][4][5] Many clinical practice and organizational guidelines recommend stopping cancer screenings when life expectancy falls below a threshold, such as 10 years. Despite guidelines and relative risks, more than 50% of adults with reduced life expectancy report ongoing cancer screening. [6][7][8] Inappropriate screening is also costly; Medicare spent $790 million on guidelinediscordant prostate, cervical, and colon cancer tests in 2009. 9 Multifactorial influences may affect cancer screenings. One conceptual model describes hierarchical factors associated with cancer care delivery. 10 Patient sociodemographic factors, insurance, and more mutable elements (eg, attitudes) may be influential. 7,11,12 Clinician-and teambased associated factors include communication and cultural norms but also roles, teamwork, and staffing. Systems-level organizational structures, community resources, and state policies also appear relevant. [13][14][15][16] Despite growing understanding, it is unclear how multilevel factors are associated with low-value cancer screenings, particularly within an integrated health system with salaried clinicians, such as the Veterans Health Administration (VHA).
The VHA uses a multidisciplinary patient-centered medical home (PCMH) model in more than 900 clinics, emphasizing teamwork, population health management, continuity, and care quality.
This model also increased capacity for preventative screenings through care registries, expanded staffing, and dedicated infrastructure. 17 Although PCMH models have been associated with improved preventative care delivery, 18,19 connections to low-value cancer screenings have not been examined. Paradoxically, more low-value screenings could occur with expanded access or protocolized screenings delivered via enhanced team structures. 17 Alternatively, low-value cancer screenings may decrease with greater clinician continuity from PCMH implementation 17 ; another study found that continuity was associated with fewer low-value tests. 20 Understanding of low-value cancer screenings is evolving. 8,9,[21][22][23] Adding to emerging evidence, validated recommendations on low-value screenings have been published. 24 Examining rates and factors associated with low-value cancer screenings within the VHA using these recommendations may provide evidence for interventions targeting the appropriateness of care or the development of low-value performance measures similar to Medicare's Merit-based Incentive Payment System. 25 Clarifying how low-value screenings are associated with PCMH organizational characteristics may inform future intervention design or implementation strategies. We sought to describe the prevalence and association of multilevel factors, including key PCMH domains, with 4 common low-value cancer screenings (breast, cervical, colorectal, and prostate) within the VHA.

Study Overview and Data Sources
We developed operational definitions of low-value screenings for each cancer (Table 1 8,9,21,22,26,27 ) from validated recommendations of ambulatory practices to avoid. 24 Those recommendations define low-value screening as outside clinical practice guidelines among patients exceeding age recommendations or with reduced life expectancy. We assessed whether patients undergoing cancer screenings had a low-value test and associated factors, among each cohort. This analysis was designated quality improvement rather than research as part of VHA primary care evaluation efforts, and thus it was not subject to institutional review board approval or waiver and was exempt from the requirement to obtain patient consent. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies. 28 We constructed 4 cohorts and examined multilevel factors using VHA administrative data: encounter data, patient and clinician demographic characteristics, and facility covariates from the Corporate Data Warehouse 29 ; staffing covariates from the Provider Specialty Workforce Report; clinic locations from the Site Tracking System; and county-level descriptors linked by patient zip code to Area Health Resource Files. 30 Race and ethnicity were defined using algorithms prioritizing self-identification. 31  Table 1. 8,9,21,26 Specific to VHA, prostate cancer screening may persist beyond guidelinerecommended age thresholds for older patients owing to higher prostate cancer risk among veterans exposed to Agent Orange, a carcinogenic tactical herbicide predominant in the Vietnam war. 34 Therefore, we conducted a subgroup analysis of Vietnam-era veterans (Ն18 years by May 7, 1975), including this exposure as a possible patient-level factor.

Cohort Criteria and Separate Low-Value Cancer Screening Outcomes
For the breast cancer screening cohort, we included women who had a mammogram, excluding those with a mammogram in 11 months prior to the index test to reduce the likelihood that diagnostic imaging followed a screening test. 26 Any patient with a high-risk diagnosis, such as breast cancer, in the prior 10 years was also excluded. 9,26,32,33 For the cervical cancer screening cohort, we included women who underwent a human papillomavirus test or Papanicolaou test, excluding those without an adequate screening history or high-risk diagnoses (eg, abnormal Papanicolaou test) in the prior 10 years 9,21,27,35 by using a validated algorithm. 21 For the colorectal cancer screening cohort, we included patients who underwent colonoscopy, sigmoidoscopy, or fecal occult blood testing or fecal immunochemical testing, excluding those with gastrointestinal symptoms in 12 months prior to the index test or with high-risk diagnoses within 14 years prior. 9,22 For the prostate cancer screening, we included men who underwent a prostate-specific antigen test, excluding those with genitourinary symptoms for 90 days prior or high-risk diagnoses in the last 10 years (eg, prostate cancer). 8,9 We also excluded African American men in accordance with other low-value literature 24 and practice guidelines recommending individualized screening decisions for African American men aged 40 to 54 years given higher prostate cancer risk. 36 We acknowledge that our use of constructed categories of race and ethnicity necessitates interpretation of our findings within the context of ancestral, structural, cultural, socioeconomic, and other factors not represented in our study.

Measurement of Low-Value Cancer Screenings
Within cohorts, we defined binary outcomes denoting whether testing met low-value criteria (Table 1). We estimated reduced life expectancy using a 1-year probability of mortality of 50% or higher, based on a validated VHA score 37 applied in prior research. 22

Potential Multilevel Factors Associated With Low-Value Care
Potential factors associated with low-value cancer screenings were based on previous studies, conceptual models, and data availability. 8

Statistical Analysis
We described unadjusted rates of low-value screenings for each cancer type. We applied exploratory logistic regression models to assess associations of multilevel factors with receipt of low-value testing among patients screened. Models excluded encounters missing patient (7%-15%) or clinic (4%-15%) covariates. We first assessed the mean estimated probability of receiving a low-value test using the marginal standardization method by sequentially adding patient-, clinic-, and clinician-level factors.
To assess independent associations from factors controlling for the others, we conducted multivariable models analyzing levels simultaneously in 2 stages owing to high proportions of missing clinician data (18%-82%

Patient and Clinical Factors
In models including both patient and clinical characteristics (Figure),    P < .001) were more likely to receive low-value tests.

Patient, Clinical, and Clinician Factors
After also adjusting for differences in ordering clinician, among patients screened for breast cancer,

Discussion
Our study is among the first to operationalize validated recommendations for low-value testing among common cancer screenings and to examine associations with multilevel factors for the VHA.
Overall, testing for breast, colorectal or cervical cancer was rarely low value, among both all averagerisk patients or screening recipients. However, low-value prostate cancer tests were more common, received by 7.6% of all average-risk men and 38.8% of screened men. Predominantly patient factors were associated with higher likelihood of receipt of low-value cancer testing among screened patients; however, no single factor was significant in one direction across all 4 cohorts. There was also no clear association between select domains of the VHA PCMH model and low-value test receipt.
The integrated system of the VHA may attract different patient populations and lead to variation in screening activities due to more equitable access, informational continuity, fewer financial barriers to care, or less incentives for services compared with other systems. 45 In illustration, in contrast to cancer screening disparities in other systems, [46][47][48]  populations. 8,54,55 Unlike other cancer screenings in which logistics may be more influential, serology-based prostate-specific antigen tests are easily obtained with other blood tests. 55 Cancer screenings may depend on patient preference, which may lead to testing outside of clinical guidelines. 56 Within VHA, we were concerned that persistent testing outside guidelines may be associated with history of Vietnam-era Agent Orange exposure, a risk factor for prostate cancer. 34 However, a dedicated subgroup analysis did not show that this exposure was associated with more lowvalue testing.
Few PCMH domains were consistently associated with low-value test receipt for any of the 4 cancers among screened patients. An association between less overall low-value care and better continuity has been described, 20 but implications for the VHA PCMH model are limited given the isolation of our findings to women screened for breast cancer.
A distinction from prior work is how we measured low-value cancer screening rates. We examined proportions of low-value testing among cancer screening recipients; other studies measured low-value cancer screening among patients at risk for low-value testing (eg, older adults). 8,9,26,57,58 For example, in 1 investigation, nearly 8% of Medicare beneficiaries older than 75 years received low-value colorectal cancer screening, whereas low-value cervical cancer screening occurred in 7% of female beneficiaries older than 65 years. 9 Another study of Medicare-enrolled women with life expectancies less than 1 year found that 18% of them still received breast cancer screening. 26

Strengths and Limitations
Defining and measuring low-value care has inherent limitations, particularly reliance on clinical practice guidelines and imperfect life expectancy predictions. A major strength of this study is our approach building from validated recommendations describing cancer screenings unlikely to improve mortality when applied to asymptomatic patients. 24 Research is needed to define the sensitivity and specificity of our claims-based measures although similar VHA-based approaches show high specificity compared with manual medical record review. 22 We did not examine factors associated with low-value testing among all eligible patients, which may make our findings more difficult to interpret. We also did not include all cancer screening factors, such as encounter time, individual attitudes, or patient request. 12,56 We also note aspects that may limit external generalizability, including specific exclusion criteria, data missingness, and studying veterans in a national integrated health system.