This is a pilot intervention study using a convergent mixed methods design. For the quantitative analysis, we will use a quasi-experimental approach, enrolling participants from an intervention site into an integrated health program and using a control site with patients not enrolled in such a program for comparison purposes. Both the intervention and control sites will be separate ambulatory care clinics managed by Michigan Medicine’s Department of Family Medicine. Both sites are located in Southeast Michigan. We will use qualitative methods to better understand social determinants of health among the patient cohort, as well as the perspectives of health care providers on the feasibility and usability of the proposed intervention. An overview of the study design is shown in Figure 1.

The intervention is based on the paradigm of integrated health programs, which are designed to improve collaboration among primary care providers, social workers, and behavioral health specialists. The co-location of these health professionals at the same clinic allows for higher levels of patient care collaboration; for example, through primary care providers regularly having regular consultations with clinical social workers. This collaboration parallels institution-wide efforts at Michigan Medicine to improve service delivery and integrate medical, social, and behavioral health in a single clinic. This model has been successfully implemented to address significant behavioral health gaps for deaf and hard-of-hearing patients in the Michigan Medicine Family Medicine clinics [13,14].

The intervention has multiple components: (1) development and implementation of an electronic health record (EHR)–based alert to improve preventive screenings among people with physical disabilities; (2) training of providers on how to improve care for people with physical disabilities; and (3) inclusion of a specialized social worker with disability health training.

We will use tailored BPAs in the EHR to provide clinical decision support to primary care providers of people with physical disabilities. This was based on a previously completed study using BPA that increased the rate of hearing screenings and audiology referrals at primary care clinics [15]. BPAs are a type of clinical decision support tool available in Michigan Medicine’s Epic medical record software called MiChart. BPAs are activated in a patient’s medical record when a patient meets eligibility criteria based on factors including diagnostic codes, procedure codes, quality indicators, or laboratory results. BPAs can be used by health care providers to more efficiently order or be reminded about screenings or immunizations. These tools can also aid in documenting medical decision-making. When this BPA is activated, the primary care provider will be prompted to engage the patient to determine the applicability of the BPA by selecting one of the recommended actions or dismissing the alert.

The BPA for this intervention is designed to prompt preventive and screening services for which there are well-documented disparities for people with physical disabilities [16-18]: hemoglobin A_1c, bone density DEXA scan, lipid panel, laboratory panels (eg, basic metabolic panel), renal ultrasound, referral to a nutritionist, and referral to a specialized social worker with disability health training. A description of the BPA trigger conditions is available in Figure 2.

When the BPA is activated for a specific patient, the alert will appear in their medical record in a yellow alert box on the main screen, along with the patient summary (left column; Figure 3). This alert remains active until it is acted on or dismissed by the provider. Available actions are “act on any of the suggested alerts,” “dismiss with an explanation” (eg, patient declines or already completed elsewhere), and “ignore.” Based on the provider’s recommendation and discussion with the patient, the provider can accept the SmartSet proposed by the BPA. The BPA will be pilot-tested by an Epic programming analyst with Michigan Medicine, TGJ, and MM before activation.

Primary care providers at the intervention clinic will be initially trained in intervention design to help reduce preventable causes of morbidity and improve functioning among people with physical disabilities. This training will be done before the implementation of the intervention and will be led by a family medicine clinician-scientist who is an expert in disability health. In addition, throughout the intervention, the research team will meet with the clinic director and other providers to communicate feedback and updates about the intervention.

Social work theory posits that an individual’s health may be influenced by organic processes as well as by psychosocial factors in the environment [19]. Therefore, social workers at Michigan Medicine have historically been well integrated with clinics [14], providing case management services, advocacy, care management for maintenance of chronic health conditions, psychotherapy, and additional social support. The social worker for this intervention will work closely with the intervention clinic’s primary care providers to provide additional support in addressing the health, behavioral, functional, and social needs of people with physical disabilities. Patients without any current identified needs will be monitored for any changes with annual BPA prompts.

Patients to be contacted by the social worker will be identified in two ways. Survey participants will complete the seven-item Generalized Anxiety Disorder Scale (GAD-7) and the Patient Health Questionnaire-9 (PHQ-9; Multimedia Appendix 1). The social worker will contact those who indicate moderate to severe anxiety or depression and assess the patient’s need for mental health resources and psychotherapy. A MiChart referral to social work will be created and prepopulated with the following common reasons for referral: caregiver support; durable medical equipment and community referrals; grief and loss; guardianship and advance directives; home modifications; long-term planning; mental health and substance abuse concerns; safety concerns; violence and abuse; and other (with a comment field). Social work interventions will mostly be completed by phone.

People with physical disabilities will be identified using the International Classification of Disease, Ninth Revision (ICD-9) and International Classification of Disease, Tenth Revision (ICD-10) diagnostic and procedure codes from patients’ problem lists in MiChart (Multimedia Appendix 2). The patient cohort at the intervention clinic will be patients who are identified as people with physical disabilities and who are patients at the intervention clinic (Briarwood Family Medicine). To create a quasi-experimental setting for our intervention, another primary care clinic in the same department will serve as a control clinic (Dexter Family Medicine).

All patients with identified physical disabilities, aged 18 years and older, will trigger the BPA. Using conservative estimates (eg, ICD-9 and ICD-10 codes with strongly associated physical disabilities in Multimedia Appendix 2), we identified 284 patients who were seen in the past 12 months before BPA implementation at the intervention clinic. The characteristics of the patients that were seen in the past 12 months at the intervention clinic were 166 (58.4%) female, 51 (18%) Black, 57 (20.1%) Medicaid insured, and 137 (48.2%) Medicare insured. A similar number of people with physical disabilities is anticipated to be identified in the control clinic.

This study will use multiple quantitative and qualitative data sources. The alignment of variables and constructs between quantitative and qualitative aims is provided in Textbox 1.

We will download structured clinical data, patient demographics, health care visits, diagnoses, procedures, laboratory orders, and results using the data warehouse at Michigan Medicine. We will also be able to capture the specific use of the BPAs for each patient. Textbox 2 summarizes the important domains that we will use in our analyses. Further, we will use an existing search algorithm [20] in the Electronic Medical Record Search Engine [21], to work with free-text and unstructured clinical documents in electronic medical records in order to identify influential social factors.

A short questionnaire (“Partners in Health”) will be prospectively collected from people with physical disabilities who meet our inclusion criteria at the time of the BPA implementation. The collected data are related to basic social needs (eg, ability to pay and social and family support; Multimedia Appendix 1).

To understand barriers and facilitators to accessing primary health care among people with physical disabilities, we will recruit people with physical disabilities who complete the aforementioned patient questionnaire and consent to being recontacted for a one-hour interview. The interview will be semistructured, guided by an interview guide (Multimedia Appendix 3), and implemented by the intervention social worker (LC) or mixed methods methodologist (TGJ).

Provider interviews will be conducted to understand usability, feasibility, and implementation constructs related to the use of the BPA alert. The interview guide (Multimedia Appendix 4) is informed by the Consolidated Framework of Implementation Research (CFIR) [22]. These interviews will be conducted by MM, a family medicine clinician-scientist, to aid in establishing rapport and trust.

Before being analyzed, the data will be explored to identify any missingness, which will be handled appropriately (eg, using imputation techniques or keeping complete cases only), based on its extent, impact on the sample size, and missing data pattern.

The characteristics of patients in the intervention group will be summarized using descriptive statistics. We will compare the characteristics of the intervention clinic with those of the control clinic and test for relevant associations using Student's t test or Wilcoxon signed-rank test for continuous variables and chi-square or Fisher exact tests for categorical variables as appropriate.

If this preliminary exploration of the full sample shows a lack of balance between the groups, we will apply inverse propensity weighting to calculate a weight defined as the inverse probability of observing the given sample [23-25].

To verify the comparability of the intervention and the control sites, which is a key requirement for our quantitative intervention effectiveness analysis, we will also analyze data on the main outcomes of interest dating back to three years before the launch of the intervention.

In addition, preliminary analyses will also include summarizing the patient questionnaire data, mental health scores (GAD-7 and PHQ-9 overall scores), and BPA usage data. The questionnaire data will be described and may be merged with EHRs to be further used in qualitative analyses. The BPA usage data analysis will be used to investigate the (1) provider type acting on the BPA, (2) type of appointment when the BPA is firing, and (3) what actions were taken with respect to the BPA order.

The primary health–based outcomes assessed will be used for preventative services (eg, diabetes and lipid screening rates and hemoglobin A_1c) and the number of emergency room or hospital visits. The primary behavioral outcomes will be based on the changes in the PHQ-9 and GAD-7 overall scores before and after the intervention (after one month).

In order to evaluate the effect of the integrated health program on these outcomes, we model these outcomes using a DD approach [26]. DD is a statistical technique applied to longitudinal data that attempts to find a causal effect between an intervention and a control group in a quasi-experimental setting before and after a certain intervention is implemented. It assumes that, if the groups were similar based on observable characteristics and preintervention trends, any differences between them with respect to the outcomes of interest can be attributed to the intervention itself, rather than to systematic, preexisting differences. The average change in the outcome in each group is compared to estimate an average intervention effect for the intervention group (known as the average treatment effect for the treated).

The key assumption DD relies on is, thus, the presence of parallel trends; in the absence of the intervention, the treatment and control groups would continue to follow analogous trends with respect to the outcomes of interest. The postintervention, parallel trend in the intervention group would be an unobserved counterfactual path. Additionally, in the context of DD, particular attention to meeting the stable unit treatment value assumption is also required; the composition of treatment and control groups should remain stable over time to avoid spillover effects [9]. Figure 4 provides a simple graphical representation of the intervention effect and the parallel trend assumption.

In the simple case of 2 time periods (1 pre- and 1 postintervention period), DD involves building a regression model with the following key predictors: a binary time covariate (before and after the intervention), a binary intervention covariate (treatment and control group), an interaction term between the 2 variables (ie, the DD estimator), and any time-varying covariates that may affect the outcome’s trend over the observed time period without being themselves affected by the intervention. Table S1 in Multimedia Appendix 5 outlines the list of outcomes and related models.

All statistical analyses will be performed using the software RStudio [27].

Qualitative data will be collected from the semistructured interviews with people with physical disabilities and primary care providers at the intervention clinic. Interviews will be audio-recorded to facilitate transcription by a Health Insurance Portability and Accountability Act (HIPAA)–compliant transcription service. Transcriptions will be imported into MAXQDA 2022 (Verbi Software) to conduct qualitative analysis.

The research team will code patient and provider interviews separately, as the codes that are developed will be different; provider interview questions are based on an implementation framework, while the patient interview questions are focused on quality of life and clinic access. The overarching qualitative strategy that will be used in this study is the qualitative description, as described by Sandelowski [28,29]. In qualitative description, the researchers seek to keep the analysis closely grounded in the participants’ words while also using theory and conceptual models as appropriate.

A preliminary codebook will be developed through multiple members of the research team open-coding the same transcript from each group and meeting to discuss coding decisions. This process encourages a collaborative process of developing codes and code definitions across multiple coders who will independently code the transcripts. In addition, transcripts from the provider group will be coded using codes aligned with the CFIR [22]. Additional codes may be added to the codebook as the analysis progresses; in this case, the qualitative analysis team will meet to discuss the appropriateness of developing a new code, the name of the code, and its definition. Previous transcripts will be rereviewed to ensure the same codebook is applied to each transcript. Throughout the analysis, coders will be encouraged to memo transcripts with overarching thoughts, theme development, and other analytic insights [30].

Themes will be developed after the coding is complete. The research team will review the developed memos, code segments, and code relations, in addition to “focusing strategies” as described by Saldana [30]. For example, we will use the “touch test” to ensure that the developed themes are conceptually relevant and not just a description of a code.