Cancer is the second leading cause of death in the United States. Health disparities exist in cancer screening and mortality. Although cancer mortality rates have decreased steadily over the past 25 years, the racial gap persists, and socioeconomic disparities are widening. Improved early detection and treatment can decrease cancer mortality, but many barriers exist to screening uptake. Cancer screening is often a trade-off for both patients and clinicians among different cancer screenings and other preventive opportunities, each of which have their own barriers to completion. This study seeks to improve quality of care and reduce health disparities in cancer mortality by using informatics to better tailor clinical discussions in order to enhance patient motivation to receive screening services and help overcome barriers to screening. The study has the following aims: 1) to develop a predictive model that identifies patients > 50 years old eligible for lung, breast and colorectal cancer screening who are at high-risk for missing these screenings; 2) to develop and pilot test a prototype shared decision-making support tool for patients and clinicians that will increase screening uptake; 3) to perform a pragmatic trial of the refined multilevel intervention that utilizes a shared decision-making support tool for patients and clinicians. During the planning phase, data from the electronic health record (EHR) will be combined with supplementary data sources (e.g. Area Health Resource File, LexisNexis) to identify patients at risk of not completing recommended screening for these cancers. Hold-out cross-validation will be used to estimate model accuracy. Machine learning model building following feature selection will utilize logistic regression and decision trees to ensure the models are easily transferable to other healthcare settings and interpretable. This predictive model will be used to develop and pilot test an application that will (A) utilize predictive modeling to identify eligible patients likely to miss screening; (B) determine, after providing patient education, patient preferences and concerns regarding screening through self-report augmented by predictive modelling; (C) identify available community resources to facilitate screening; (D) provide this information to patients and to clinicians in the EHR prior to clinical visits to assist shared decision-making conversations and support active problem solving of identified barriers. This multilevel intervention will be pilot-tested and refined within 4 academic primary care practices over 6 months (N=24 clinicians; 8181 potential screenings yearly). A pragmatic trial will be conducted during the R33 phase using a stepped wedge cluster randomized design (N=45 clinicians; 11,239 potential screenings yearly). Completed cancer screening will be the primary outcome. Patient experience, clinician satisfaction and utilization will also be assessed. If successful, this approach will lead to an intervention tailored to the patient and their community that leverages information technology in a scalable way that is transferrable to other clinical settings and types of preventive services.
Significant disparities exist in breast, colorectal and lung cancer, the leading causes of cancer mortality, requiring the development of new interventions that increase screening uptake in diverse populations. We propose an intervention that employs a multilevel approach that leverages the Electronic Health Records (EHR) and supplementary data sources to develop a predictive model to identify eligible patients at high-risk of not obtaining screening and to facilitate improved shared decision-making conversations using an EHR based decision-making support tool that interacts with patients and clinicians. By personalizing the approach to screening and improving the quality of patient-clinician interactions, we hope to enhance patient motivation to receive screening services and reduce health disparities.
