ASPO Abstracts
It's not just neighborhoods: Multiple levels of influence on cervical cancer screening across multiple healthcare settings
Category: Early Detection & Risk Prediction
Conference Year: 2020
Abstract Body:
Purpose: Describe variability in cervical cancer screening across multiple cross-classified
levels: healthcare
provider, clinic, and neighborhood of residence.
Methods: We examined multilevel variation in Pap screening among patients due for
screening using electronic
health record (EHR) data from 3 healthcare settings in the PROSPR network. This
retrospective cohort study
included average-risk women aged 21-65 who completed ≥1 clinical encounter, had 3
years of EHR data, and
whose addresses were linked to census tract data. We excluded women who were:
HIV+, already screening
up-to-date, not screening-eligible, or with history of cervical cancer or prior cervical
abnormality. The outcome
was Pap screen within 3 years of cohort entry (yes/no). We fit Bayesian cross-classified
4-level logistic models
nesting patients within separate, non-overlapping levels (providers, clinics, census
tracts). We estimated
multilevel variation using Median Odds Ratios (MOR) and 95% credible interval in empty
models (i.e., no
covariates) and after adjusting for patient-level covariates (age, sex, race/ethnicity, BMI,
pregnancy and
smoking status, number of healthcare encounters, and cohort entry year).
Results: Approximately half of eligible patients were screened in each setting (1: 54.4%
of n=81,668; 2: 48.4%
of n=42,127; 3: 47.3% of n=43,991). The magnitude of variation in Pap among all levels
was similar among
settings. After adjusting for patient-level covariates, the largest variation was at the clinic
level (MOR range:
1.44-1.62), followed by provider (MOR range: 1.27-1.35), and neighborhood level (MOR
range: 1.17-1.21). For
example, we can interpret clinic-level MOR as: if a woman switched from a low-screening
to a high-screening
clinic, her odds of Pap screen would be 1.44-1.62 times higher. Future results will include
one additional
healthcare setting and will include provider-, clinic-, and neighborhood-level covariates.
Conclusions: Significant and substantial variability in Pap was observed across provider,
clinic, and
neighborhood levels in 3 different healthcare settings, suggesting that factors at multiple
levels are driving
cervical cancer screening. Future interventions must identify and modify multilevel
influences to reduce
disparities in Pap screening.
Keywords: multilevel, cervical cancer screening, disparity, variation