“Patient Lost to Follow-up”: Opportunities and Challenges in Delivering Primary Care in Academic Medical Centers

Abstract

Background

Academic health centers (AHCs) face unique challenges in providing continuity to a medically and socially complex patient population. Little is known about what drives patient loss in these settings.

Objective

Determine physician- and patient-based factors associated with patient loss in AHCs.

Design

Retrospective cohort study, embedded qualitative analysis.

Setting

Academic health center.

Participants

All visits from 7/1/2014 to 6/30/2019; 89 physicians (51%) participated in a qualitative analysis.

Measures

Physician-based factors (gender, years of service, hours of practice per week, trainee status, and departure during the study period) and patient-based factors (age, gender, race, limited English proficiency, public health insurance, chronic illness burden, and severe psychiatric illness burden) and their association with patient loss to follow-up, defined as a lapse in provider visit greater than 3 years.

Results

We identified 402,415 visits for 41,876 distinct patients. A total of 9332 (22.3%) patients were lost to follow-up. Patient factors associated with loss to follow-up included patient age < 40 (HR 3.12 (2.94–3.33)), identification as non-white (HR 1.07 (1.10–1.13)), limited English proficiency (HR 1.18 (1.04–1.33)), and use of public insurance (HR 1.12 (1.04–1.21)). Provider factors associated with patient loss included trainee status (HR 3.74 (2.43–5.75)) and having recently departed from the practice (HR 1.98, 1.66–2.35). Structured interviews with clinical providers revealed unfavorable relationships with providers and staff (35%), inconvenience accessing primary care (23%), unreliable health insurance (18%), difficulty accessing one’s primary care provider (14%), and patient/provider transitions (10%) as reasons for patient loss.

Conclusions

Younger patient age, markers of social vulnerability, and physician transiency are associated with patient loss at AHCs, providing targets to improve continuity of care within these settings.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11606-021-07216-3.

INTRODUCTION

Academic health centers (AHCs) play a vital role in the delivery of primary care in the USA, fulfilling the tripartite mission of access to excellent clinical care, advancing research and innovation, and training the next generation of physicians. In doing so, they serve a dually complex patient population, with significantly higher medical and social complexity compared to community-based health centers (CHCs) ^1,2. However, compared to CHCs, AHCs have lower rates of patient continuity ². Continuity in patient care is a critical component to providing high-quality primary care, not only because it prevents unnecessary healthcare costs ^3–5, but it also reduces morbidity and mortality ⁶. Therefore, understanding the factors that lead to patient loss will be essential to the provision of high-quality primary care within AHCs.

While some patient loss is likely due to non-modifiable factors, such as death and patient migration, little is known about the patient- and physician-based factors that drive loss to follow-up in primary care. Most studies were conducted within CHCs or in specific patient populations ^7–9, which have found that the increased medical complexity, female gender, and older age are predictors of patient retention ⁷, and that higher degrees of social complexity and mental illness are predictors of patient loss ^8,9. However, little is known about the factors that may drive patients to leave AHCs. Determining such factors will be critical to improving the value of care provided by these sites.

We aim to determine significant drivers of patient loss within a large urban hospital-based primary care clinic, to better inform future interventions within AHCs. We hypothesize that similar to CHCs, younger patient age and social vulnerability are factors associated with patient loss. We also hypothesize that there are additional factors related to the unique characteristics of AHCs, specifically, their high rates of patient and physician transitions.

METHODS

Setting and Participants

This is a retrospective, single-center cohort study set in a large academic hospital-based primary care clinic where 60 faculty preceptors and 110 residents provide primary care on multidisciplinary teams to 41,876 patients in a large metropolitan area. Residents serve as the primary care physician for 7873 of these patients, supervised by faculty. We included all visits to the practice between 7/1/2014 and 6/31/2019.

Data Sources

We derived our cohort and variables of interest using the hospital’s clinical, administrative and professional fee billing repositories, and internal provider-linked registries. These data are prospectively collected as part of regular hospital operations. Please see Appendix 1 for the source and description of each variable in our model.

Outcomes

Primary Outcome

Our primary outcome was patient loss from our primary care clinic, defined as a lapse in visits to the practice of greater than or equal to 3 years during the study period. For each patient, we set the index clinical visit as the earliest date of service. Each subsequent date of service provided a time interval between clinic visits. Loss to follow-up occurred if there was a lapse greater than or equal to 3 years, and the time to loss to follow-up is defined as the time interval between the index visit and the last visit before loss to follow-up plus 3 years.

Covariates

We grouped variables into two conceptual categories: provider-based variables and patient-based variables. Please see Appendix 1 for each variable’s source and how it was operationalized.

Provider-based variables

Each date of service is linked to a provider identification number for the patient’s primary care physician. By linking this to an internal registry, we were able to derive the following variables for each patient’s primary provider: trainee vs. faculty status, gender, years of service at our institution, number of clinical practice hours per week, and whether the provider had left the practice during the study period.

Patient-based variables

We were interested in measuring the association of both psychosocial and medical factors with patient loss. We created variables for patient age, with attention to dichotomizing patients who were younger or older than 40 years of age; sex, race, primary language, and public vs. private insurance type. Each of these variables are linked to the date of service through the patient’s medical record number. To adjust for a patient’s medical complexity, we linked the first nine billing codes for each date of service using our billing data repository and calculated a Charlson comorbidity score. Finally, we used these diagnosis codes to adjust for major psychiatric co-morbidities.

Analysis

Patient characteristics were summarized with a mean and standard deviation (for continuous variables) or percentages (for categorial variables). We fitted a Cox regression model and derived hazard ratios on patient- and provider-based determinants of loss to follow-up. Because resident and faculty cohorts differed in both patient and provider characteristics, we conducted our analysis on the whole population of patients, and then on the faculty and resident-only population separately. All statistical analyses were performed using SAS version 9.4 (Cary, NC).

Embedded Qualitative Analysis

In order to study provider perceptions of why patients are lost to follow-up in primary care, we created a three-question, anonymous online survey informed by discussions with our patient family advisory committee and discussions with primary care leadership (Appendix 2 Fig. 1). The survey was sent out electronically to all physicians in the practice by a member of the research team (RM). Participation in the survey was voluntary and was not incentivized. We utilized pertinent sections of the COREQ checklist for reporting and analyzing provider survey responses. We used thematic analysis to identify major recurring themes within provider narratives.

Ethical Issues.

The Institutional Review Board at Beth Israel Deaconess Medical Center approved the protocol as exempt from further review as an educational research project. Participation by individual physicians was voluntary and all data were kept confidential.

RESULTS

We identified a total of 41,876 patients with visits to the practice from July 1, 2014, to June 31, 2019, of which 9332 (22.3%) patients experienced loss to follow-up during the study period. In an unadjusted descriptive analysis, compared to patients who were retained in the practice, patients who were lost to follow-up were younger (mean age 41 vs. 53.7 years, standard mean difference = 0.66), and their primary providers were more likely to be a physician-in-training (32.4 vs. 14.9%, standard mean difference = 0.46), or a faculty member who recently left the practice (14.5 vs. 7.9%, standard mean difference = 0.29).

In the adjusted analysis of our entire patient cohort, factors associated with loss to follow up were patient age < 40 years (HR 3.12; CI: 2.94–3.33), identification as non-white (HR 1.07; 1.10–1.13), limited English proficiency (HR 1.18; 1.04–1.34), and the use of public insurance (HR 1.12; 1.04–1.21). Patients were also more likely to be lost if they had a trainee as their primary provider (HR 3.74, 2.43–5.75). However, patients were less likely to be lost if they were female (HR 0.93, 0.97–0.99), had higher medical complexity based on a Charlson comorbidity index (HR 0.87; 0.85–0.89), or had a history of severe psychiatric illness (HR 0.68; 0.52–0.90).

We repeated our model within a faculty and trainee-only provider population. For both faculty and trainee physician populations, patient age < 40 years (HR 2.85, 2.65–3.06; HR 2.82, 2.48–3.21) was significantly associated with patient loss; and higher medical complexity (HR 0.87 (0.85–0.90); HR 0.86 (0.81–0.91)) was significantly associated with patient retention. Within the resident patient cohort (and not in the faculty cohort), limited English proficiency (HR 1.29, 1.04–1.60) and provider male gender (HR 1.48, 1.03–2.13) were significantly associated with patient loss, while severe psychiatric illness was significantly associated with patient retention (HR 0.51, 0.29–0.90). In the faculty patient cohort having a provider who had recently transitioned out of the practice (HR 1.98, 1.66–2.35) was significantly associated with patient loss.

We surveyed providers at our primary care practice to get their perspectives on reasons for patient loss (see Supplement 1 for a copy of our survey). We received a total of 89 responses. In total, 46 of 89 (52%) were faculty physicians (a 70% response rate); 43 of 89 (48%) were trainees (a 39% response rate). When asked to identify the main drivers of patient loss in primary care, we noted the following themes for responses: unfavorable relationships with providers and staff (35%), inconvenience accessing primary care (23%), health insurance changes (18%), difficulty accessing one’s primary care provider (14%), and patient/provider transitions (10%).

DISCUSSION

In this large academic hospital-based primary care clinic caring for nearly 42,000 patients, we found that a significant percent (22.3%) of patients were lost to follow up during the 5-year study period. Factors that are unique to the clinical environment of academic medical centers (AMCs), namely, younger patient age, physician transitions, and markers of social vulnerability, emerged as significant associations with patient loss, while medical and psychiatric complexity were found to be protective. These findings highlight the unique opportunities and challenges of delivering primary care within AMCs (Tables 1, 2 and 3).

Table 1.

Patient Demographics by Lost vs. Retained Status

	Retained (n = 32,544, 77.7%)	Lost (n = 9332, 22.3%)	Standard mean difference
Age < 40 years (%)	23.1	55.8	0.66
Sex (% female)	58.1	56.7	2.80
Race (% white)	44.7	43.1	0.03
Limited English proficiency (%)	6.3	5.8	0.02
Public health insurance (%)	15.8	17.1	0.05
Provider is a trainee (%)	14.9	32.4	0.46
Provider recently left the practice (%)	7.9	14.5	0.29
Provider gender (% female)	47.9	44.7	0.06
Provider years of service (mean, sd)	15.8	16.1	0.03
Cancer screening (% screened)
Breast cancer	85.3	64.7	0.60
Cervical cancer	82.9	75.1	0.22
Colorectal cancer	87.6	66.9	0.66

Table 2.

Patient and Provider-Based Factors Associated with Patient Loss at an Academic Medical Center

Entire cohort	N = 42,826 HR (95% CI)1
Patient characteristics
Female gender	0.93 (0.87–0.99)
Age < 40	3.13 (2.94–3.33)
Nonwhite	1.07 (1.01–1.13)
Limited English proficiency	1.18 (1.04–1.34)
Public health insurance	1.12 (1.04–1.21)
Medical complexity2	0.87 (0.85–0.89)
Severe psychiatric illness3	0.68 (0.52–0.90)
Provider characteristics
Male gender	1.30 (0.92–1.84)
Trainee status	3.74 (2.43–5.75)

¹We modeled patient loss, defined as a visit interval of greater than or equal to 3 years, using a Cox regression model and the above patient- and provider-based covariates

²Medical complexity was measured using a Charlson co-morbidity index

³Defined as having bipolar disorder or a psychotic disorder

Table 3.

Predictors of Patient Loss, Stratified by Physician Type

	Faculty cohort n = 34,020, HR (95%CI)1	Resident cohort n = 7856, HR (95%CI)1
Patient characteristics
Female gender	0.93 (0.86–1.01)	0.95 (0.85–1.05)
Age < 40	2.85 (2.65–3.06)	2.82 (2.48–3.21)
Nonwhite	1.06 (0.99–1.14)	1.02 (0.92–1.13)
Limited English proficiency	1.07 (0.91–1.25)	1.29 (1.04–1.60)
Public health insurance	1.09 (0.99–1.21)	1.00 (0.88–1.14)
Medical complexity2	0.87 (0.85–0.90)	0.86 (0.81–0.91)
Severe psychiatric illness3	0.78 (0.58–1.06)	0.51 (0.29–0.90)
Provider characteristics
Male gender	1.09 (0.76–1.56)	1.48 (1.03–2.13)
Provider clinical hours per week (per hour)	0.99 (0.97–1.02)	-
Provider years of service (per year)	1.03 (1.01–1.05)	-
Provider recently left the practice	1.98 (1.66–2.35)	-

¹We modeled patient loss, defined as a visit interval of greater than or equal to 3 years, using a Cox regression model and the above patient- and provider-based covariates

²Medical complexity was measured using a Charlson co-morbidity index

³Defined as having bipolar disorder or a psychotic disorder

Across both trainee and faculty patient populations, younger patients (defined as under the age of 40 years old) were nearly three times more likely to be lost to follow-up when compared to their older counterparts. This finding is pertinent to AMCs, which are usually located in large metropolitan areas that attract a younger, more transient population. These patients and their unique needs may not be best served by a classic model of primary care delivery, which is designed around chronic disease prevention and management, and patient-provider continuity. However, patients under 40 years of age still have significant healthcare needs, specifically, urgent care, mental health, sexual and reproductive health, cervical cancer screening, and vaccination among many others. Re-designing preventive healthcare for younger patients that enables accessible urgent care platforms (including virtual platforms), mental healthcare, and access to centralized repositories for needed preventive care would be mutually beneficial for patients and their physicians.

Physician transition also emerged as being strongly associated with patient loss, specifically as faculty members leave the practice and trainees graduate from training programs. This is yet another unique feature of AMCs where a subset of faculty may view the AMC as part of their early faculty development, eventually seeking more permanence away from busy metropolitan hospital centers. Furthermore, trainee graduation and transition is an inevitable event each academic year, as AMCs are built upon the training mission. This indicates the need for robust systems to ensure patients are retained during a physician transition at AMCs. Such systems could include utilizing clinical teams rather than one-on-one physician–patient relationships, and robust empanelment strategies embedded into population health management.

Markers of social vulnerability, including underrepresented minority status, limited English proficiency, and carrying public health insurance, were all associated with patient loss. When this analysis was repeated in each physician population, the effect only persisted in the trainee population, where physician male gender also emerged as significant. Notably, prior works at our institution and other AMCs locally have revealed significant differences in the psychosocial illness burden between trainee and faculty patient populations ^10,11; therefore, the difference in the strength of the association between physician population may be related to statistical power rather than a physician effect. Regardless, our findings point to the urgent need to build structurally equitable systems to retain the most vulnerable patients, particularly within AMCs where the majority of underserved care takes place.

Finally, individuals with higher medical and psychiatric illness burden were significantly less likely to leave the practice. This may point to a high quality of care provided to these patients or highlight that our current models of primary care are designed around management of chronic illnesses, and that AMCs specialize in the care of medically complex patients, and therefore naturally retain these patients. Alternatively, it may point to high barriers for departure for these patients due to higher reliance on their providers for medically complex patients and issues of agency for socially complex patients.

Our qualitative results provide a unique perspective on patient loss from the provider’s perspective. Physicians identified unfavorable relationships within the healthcare system most often, followed by a subset of challenges related to the logistics of primary care, including the inconvenience of physically getting to the practice, sudden changes in medical insurance coverage, and a system that limits patient access to their physician. Taken together, there is a theme of a lack of patient-centeredness driving patient loss in primary care—both from a relational and logistical perspective.

Our study calls into question how we define a primary care panel. It is noteworthy that age less than 40 years was the strongest predictor of loss, and that chronic disease burden seemed to be protective. This may simply underline that different patients may in fact need different intervals between routine healthcare. Most practices use a lapse in care of 2–3 years as their internal definition of active panel, as there is no standard way to measure this. Our study suggests the definition could vary based on patient age and comorbidities, and that perhaps a young healthy patient with few preventive care needs could see their PCP every 5 years and still be considered empaneled whereas a medically complex patient in their 70 s should be considered lost to follow-up if they have not seen their PCP in under 1 year.

Our study has several limitations. First, we did not have a sufficient number of non-white providers to pursue an analysis on racial or ethnic concordance between patient and provider, which could be a factor that promotes retention. Second, we calculated Charlson comorbidity indices based on ICD codes. It is possible that these codes may not fully capture a patient’s true morbidity, as providers may not bill for every diagnosis a patient carries. Finally, patient insurance status was identified based on the health insurance they had at their last encounter with our practice. Some patients may have experienced health insurance changes subsequent to that encounter that may have limited their ability to continue to receive care at our primary care practice depending on their provider network. As there is no way for us to prospectively identify changes in patients’ health insurance status within our medical record system alone, we were unable to capture these changes. Finally, while our qualitative study allowed us to gain a deeper understanding of provider perceptions on patient attrition, we were not able to survey patients due to significant challenges in contacting patients who had left the practice.

In summary, we found that the unique characteristics of the patient and physician populations within academic medical centers are significantly associated with risk of patient loss. Our qualitative study results aligned with our quantitative results, demonstrating that providers are aware of the factors that tend to lead to patient loss, adding to the internal validity of our study results. Among these, younger patient age, attending physician departure and trainee status, and several social determinants of health each require a specific approach. Building robust team-based care models and empanelment strategies both provide more secure connections for patients to their health centers and may stand in for the traditional physician–patient continuity relationship in a setting where continuity is often disrupted by the transience of the populations of providers and patients at AMCs. Adapting our models to increase structural equity and to engage younger patients will also address some of the unique needs we face. Our findings highlight the need for innovative primary care delivery. Perhaps AMCs, with their academic mission, are uniquely positioned to lead this work.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Funding

This work was conducted with support from Harvard Catalyst | The Harvard Clinical and Translational Science Center (National Center for Research Resources and the National Center for Advancing Translational Sciences, National Institutes of Health Award UL1 TR002541 and financial contributions from Harvard University and its affiliated academic health care centers. The content is solely the responsibility of the authors and does not necessarily represent the official views of Harvard Catalyst, Harvard University and its affiliated academic health care centers, or the National Institutes of Health.

Declarations

Conflict of Interest

The authors declare that they do not have a conflict of interest.