In this population-based case-control study, we compared cervical cytology screening histories of women who died of cervical cancer (cases) to those of a sample of women at risk of cervical cancer who were otherwise similar (controls) from 2 health plans in the US Pacific Northwest. Receipt of cervical cancer screening during the presumed detectable preclinical phase (DPP) of cervical cancer development was the primary exposure (bolded along the x-axis in Figure 1. For cervical cytology screening, the DPP begins when a premalignant lesion is detectable by cytology and ends with the onset of clinical signs or symptoms of ICC. A deficit of screening among cases relative to controls should be observable during this period if the test is beneficial. Inclusion in the analysis of periods prior to or after the DPP attenuates the estimated benefit.
Screening during the detectable preclinical phase (DPP) of persons who died of their cancer (cases) and during the corresponding period among controls. The DPP is shown in bold along the x-axis. This figure assumes that the DPP is 7 years for all cases, that screening during the first 6 years of the DPP inevitably leads to the cancer being cured (i.e., odds ratio = 0), and that screening after that time is of no benefit to survival. Adapted from Weiss et al. ().
Operationally, we first ascertained for each case the date of onset of clinical signs or symptoms that led to the diagnosis of cervical cancer (the index date) by means of a standardized medical record review. Cervical cancer signs or symptoms were defined as postmenopausal bleeding, postcoital bleeding, vaginal bleeding, nonspecific bleeding, abdominal pain, vaginal discharge, weight loss, obstructive uropathy, or ascites in the 12 months prior to cervical cancer diagnosis. If a case was screen detected, the screening date was the index date. The index date for each control matched that of her respective case. Second, whether a subject was screened for cervical cancer in the 7 years prior to the index date was ascertained from the medical records. Only screening tests (i.e., those performed in the absence of cervical cancer signs or symptoms) that occurred within 7 years before the index date were included in the analysis, because ICC incidence among Pap smear–negative women aged 55–79 years returns to that of unscreened women 5–7 years after a negative test. Numerous studies have corroborated this 7-year estimate of the DPP for cervical cancer, though others have estimated that cervical cancer is detectable for up to 30 years. If the true DPP began more than 7 years before the index date—that is, if cervical cancer precursors were detectable by Pap smear for more than 7 years prior to the onset of cervical cancer signs or symptoms—then undercounting of screening Pap smears would be relatively greater among controls; controls' screening is expected to have been relatively uniform during the case's DPP, whereas screening tests among cases would likely have occurred just prior to diagnosis, at the end of the DPP (Figure 1). Thus, if the presumed DPP was shorter than the true DPP, any difference in screening prevalence between cases and controls would be minimized, and the calculated odds ratio would underestimate the benefit of screening.
Study subjects were identified from enrollees in 1 of 2 integrated health plans: Group Health, based in Seattle, Washington, and Kaiser Permanente Northwest (KPNW), based in Portland, Oregon, which together cover nearly 1.1 million individuals at present. Though these health plans' screening algorithms recently incorporated human papillomavirus (HPV) DNA testing to triage equivocal cytology results, the result of an HPV DNA assay did not influence whether cytology was performed during any portion of the study period. Because Pap smears were administered independently of HPV DNA testing, the use of HPV DNA testing does not undermine this study's ability to ascertain the benefit of cytology screening.
Cases were women who died of cervical cancer during 1980–2010 at ages 55–79 years. Cases were ascertained from the Cancer Surveillance System for Group Health enrollees (part of the National Cancer Institute's Surveillance, Epidemiology, and End Results program) and from the Kaiser Tumor Registry for KPNW enrollees.
Controls were a sample of women aged 55–79 years enrolled in Group Health or KPNW during 1980–2010. To be a match, a control must have been enrolled in the health plan on her case's tumor registry diagnosis date (reference date). Matching was based on health plan, age (within 6 months), and duration of health plan enrollment prior to reference date (equal to or greater than that of the case by no more than 6 months). For both cases and controls, eligibility was restricted to women with 6 or more years of enrollment prior to the reference date, with no gaps in enrollment greater than 6 months. For Group Health controls, eligibility was restricted to women residing in the 13 western Washington counties surveyed by the Cancer Surveillance System. Eligible controls were ordered in terms of how well they were matched to the case, measured as the sum of the number of days between the case's and potential control's 1) birth dates and 2) health plan enrollment dates. The 2 potential controls with the lowest sums of these 2 numbers were selected for medical record review. Upon review, any potential control who had a hysterectomy prior to the reference date or who had evidence of receiving any health care outside the health plan was excluded and replaced with the next best-matched control, until 2 eligible controls with intact cervices were identified per case.
Cervical cancer screening history was recorded from medical records using a standardized medical record abstraction database developed with Microsoft Access, version 11.5, software (Microsoft Corp., Redmond, Washington). The reasons for the test, whether diagnostic or screening in nature, were ascertained for each Pap smear; only screening tests were included in the analysis (i.e., those performed in the absence of ICC signs or symptoms). Data on potential covariates were obtained from the medical record, including marital status, body mass index (weight (kg)/height (m)), smoking history, race/ethnicity, parity, menopausal status, oral contraceptive use, and immunosuppressive status. Data were not available on sexual history or HPV infection. The institutional review boards at Group Health and KPNW each approved the study protocol.
Multivariate logistic regression was used to quantify the odds ratio of cervical cancer death associated with screening during the DPP, adjusting for matching variables and covariates that were associated (P < 0.10) with case status. Unconditional logistic regression was used because matching variables were easily quantified and conditional logistic regression would have decreased study efficiency. Because the observed effect of screening may differ with estimated DPP length, sensitivity analyses were performed by varying the DPP duration in 6-month intervals from 5 to 7 years prior to the index date. Three exploratory analyses were planned: 1) stratification by age at diagnosis into less than 65 years or 65 years or older, the age at which the US Preventive Services Task Force (Rockville, Maryland), the American Cancer Society (Atlanta, Georgia), and the American Congress of Obstetricians and Gynecologists (Washington, DC) recommend cessation of screening for most women, 2) stratification by year of diagnosis (1999 or earlier vs. 2000 or later), because a major advance in cervical cancer treatment—the addition of chemotherapy to radiation therapy–based treatment—occurred in 1999, and 3) restriction to cases with squamous cell carcinoma (SCC), because cytology screening may be more efficacious in preventing SCC than adenocarcinoma.
All statistical tests were 2-sided. The primary data analysis software used was Stata, version 12.0 (StataCorp LP, College Station, Texas).