Abstract
Purpose: To determine whether an immediate referral to a medical-legal partnership (MLP), compared with a 6-month waitlist control, improved mental health, health care use, and quality of life.
Methods: This trial randomly assigned individuals to an immediate referral or a wait-list control. The MLP involved a collaboration between the primary care clinic and a legal services organization. The primary outcome was stress (6 months) as measured by the Perceived Stress Scale (PSS). Secondary measures included the Center for Epidemiologic Studies Depression Scale; Generalized Anxiety Disorder scale (GAD-7); Patient-Reported Outcomes Measurement Information System (PROMIS); and emergency department (ED), urgent care, and hospital visits. Assessments were at baseline and 3-, 6-, and 9-month follow-ups. Bayesian statistical inference and a 75% posterior probability threshold were used to identify noteworthy differences.
Results: Immediate referral was associated with lower PSS scores and higher GAD-7 scores. PROMIS scores were higher for the immediate referral group with respect to several subdomains. At 6 months, the immediate referral group demonstrated 21% fewer ED visits and 75.6% more hospital visits.
Conclusion: Immediate referral to the MLP was associated with lower stress and a lower rate of ED visits but higher anxiety and a higher rate of hospital visits.
Trial Registration: ClinicalTrials.gov Identifier: NCT03805126.
- Clinical Trials
- Family Medicine
- Health-Harming Legal Needs
- Mental Health
- Primary Health Care
- Quality of Life
- Referral and Consultation
- Social Determinants of Health
Introduction
Social risk factors contribute to undesirable health outcomes,1⇓⇓⇓⇓⇓–7 leading to calls to connect social and medical care.8,9 Spurred by programs such as Accountable Health Communities, primary care clinics are redesigning clinical workflows to identify individuals with social needs and refer them to community service providers.9–11 One promising intervention is the medical-legal partnership (MLP), which coordinates services between medical and legal organizations to address health-harming legal needs (HHLNs).12,13 The evidence base to support MLPs is growing,12,14 with studies demonstrating that MLPs stabilize housing, enhance finances, and reduce stress.15⇓⇓–18 Unfortunately, this body of work has been limited by the absence of randomized-controlled trials (RCTs).17 Thus far, 2 studies, in pediatric populations, randomized participants to legal services and concluded that MLPs increased preventive services, reduced emergency department (ED) visits, and improved diabetes control.17,19
In 1967, an attorney worked at the nation’s first community health center,20 and nearly 30 years later, medical and legal professionals in Boston collaborated to help children with asthma. As this model expands, there is a need to evaluate MLPs nationwide.21 Seeking to address social needs, a legal services organization and academic partner launched an MLP in 2018. The objective of this trial, the first of its kind in adults, was to assess whether referrals to the MLP were associated with improvements in stress (primary outcome), anxiety, depression, health care use, and quality of life and to identify the services provided by the MLP.
Methods
Study Settings and Participant Eligibility
The study took place from February 14, 2019, until September 30, 2020, in an urban primary care clinic. Eligible participants were low income (earning less than 200% of the federal poverty level), 18 years of age or older, spoke English or Spanish, and screened positive for HHLNs.
Study Procedures, Randomization, and Follow-Up
Individuals were screened using an instrument developed by the authors (Appendix). Those who identified as being at significant and immediate risk were excluded from randomization, directly referred to the MLP, and not included in analyses. To identify these individuals, we asked this question: “In your opinion, do your legal needs pose a significant and immediate (which means in the next several days) risk of serious personal harm to you or your close family members? For example: an eviction warning or a court date.” Randomization was achieved using a computer-generated algorithm, with variable block sizes of 4 and 6, and 1:1 allocation to immediate referral to the MLP or placement on a 6-month waitlist control. The randomization sequence was generated by the study’s statistician and implemented through REDCap by a research assistant. Three research assistants conducted in-person, baseline assessments and in-person, online, or telephonic 3-month, 6-month, and 9-month assessments (English or Spanish). Participants were entered into 2 $100 gift card drawings if they completed the 6-month and 9-month assessments, respectively.
Intervention
Those randomized to immediate referral were referred on the same day, whereas those on the waitlist were referred 6 months after randomization. The legal services organization made multiple attempts via telephone (typically 5) and mail. If successful, they conducted intake for eligibility and identification of HHLNs. Due to funding restrictions, the legal services organization could not accept cases that were out of state; involved conflicts of interest; or pertained to immigration, personal injury, or medical malpractice. In these situations, a list of resources was provided. For accepted cases, an attorney and paralegal delivered advice and counsel (both in person and over the phone), drafted documents, and provided legal representation at no cost. The paralegal was physically at the clinic 1 to 2 times per month. All had access to therapists, a social worker, and a community health worker at the clinic.
Measures
The Perceived Stress Scale (PSS) assessed the extent to which respondents found their lives to be unpredictable, uncontrollable, or overloading. The PSS at 6 months served as our primary outcome and primary endpoint. Higher scores (0 to 40) indicate greater psychological stress and correlate with mental and physical exhaustion.22,23 The Center for Epidemiologic Studies Depression Scale (CES-D) measured depressive symptoms.25 Higher scores (0 to 60) indicate more depressive symptoms and correlate with other measures of depression.26,27 The Generalized Anxiety Disorder scale (GAD-7) assessed anxiety. Higher scores (0 to 21) correlate with lower functional status and disability.28
To assess quality of life, we administered the 29-item Patient-Reported Outcomes Measurement Information System (PROMIS) instrument (Profile version 2.1).29 This instrument assesses physical function, anxiety, depression, fatigue, sleep disturbance, ability to participate in social roles and activities, pain interference, and pain intensity. Raw scores were transformed to T-scores (0 to 100; mean of 50 and SD of 10 for the US general population).30 The T-score rescales raw scores into standardized scores with a mean equal to 50 and SD equal to 10 in reference to a population. Higher scores represent more of the domain being measured.31,32 Thus, higher fatigue scores indicate more severe symptoms of fatigue.
To assess health care use, participants reported the number of urgent care, ED, and hospital visits over the prior 6 months (at baseline) and during the time since the previous assessment. Demographic data (age, sex, race, ethnicity, and language) were extracted from the electronic health record.
Given the lack of a universally accepted HHLN screening tool, we collaborated with researchers and legal professionals to develop one33⇓–35 and obtained permission when needed.36,37 The screening tool (Appendix) is 25 items and encompasses legal issues, including income, insurance, safety, guardianship, housing, and food. We screened all eligible participants because no optimal screening strategy for HHLNs exists, and patients may not know they have HHLNs. We summed HHLNs via the screening instrument, by adding the positive responses for each subdomain. Since respondents could select multiple answer choices on paper, positive responses to the questions regarding housing instability, utilities, and food insecurity (separately for low and very low) were only counted once.
Receipt of Legal Services
Because of attorney-client privilege, the legal services organization could only share data for those participants who chose to sign a release. Among those who did, we tracked the HHLNs identified and addressed by the legal services organization. After completing an intake form and interview, legal professionals recorded whether they accepted or rejected (eg, they failed to respond or meet eligibility criteria) cases and the subsequent outcomes of those cases.38
To determine whether participants received legal services (a step needed for the per-protocol analysis below), we first categorized those whose cases were accepted as receiving legal services and those whose cases were rejected as not receiving legal services. Then, among the remaining participants, we categorized participants as not receiving legal services if they reported that they had not communicated with the MLP (a question asked during the 3-, 6-, and 9-month assessments) or if they did not have a record in the data management system of the legal services organization. We also report the categories of legal services provided.
Sample Size
Prior interventions using the PSS have reported mean differences of approximately 3 (scale range: 0- to 40; SD = 6).23,24,39,40 Given a 2-sided α of 0.05 and β of 0.2, a priori power estimates determined we needed to recruit at least 64 people per group (128) to detect this difference. Accounting for loss to follow-up (20%), we needed to randomize 80 per group (or 160 total).
Blinding
Participants were not blind to their group. The research assistants completed the assessments and were blind to the group assignment with notable exceptions. For example, participants, on occasion, discussed their interactions with the MLP, revealing their group to the research assistants. In these cases, the research assistants were unblinded when conducting subsequent assessments. Furthermore, 1 research assistant sent the referrals and therefore was also not blinded. This individual primarily conducted the baseline and 9-month assessments (after all had received referrals). We conducted analyses blind to treatment group.
Statistical Analyses
Descriptive statistics were used to evaluate characteristics. Analyses were performed on an intention-to-treat sample of eligible participants. Generalized linear modeling was used to model the outcomes (6 months), controlling for baseline levels of the outcome. Before testing each outcome, characteristics were screened as potential confounders of the relationship between outcome and group (primary analyses: randomized groups; per-protocol analyses: received intervention groups). To test whether variables demonstrated a relationship with the group and outcome,41,42 Chi-square and Mann-Whitney-Wilcoxon tests were used to screen for confounding (see Table 1 for P values). No variables met criteria for confounding in either analysis (ethnicity demonstrated a relationship with the received intervention grouping variable, but none of the outcomes).
Participant Characteristics of a Randomized Controlled Trial Testing a Medical-Legal Partnership Referral, by Total Sample, Randomized Condition, and Whether Participants Received Legal Services
The analyses of health care use measures relied on the negative binomial distribution to handle overdispersed count outcomes, and the resulting regression coefficients were exponentiated to provide rate ratios (RRs). The psychological measures’ total scores were modeled as binomial proportions of maximum possible values. As a result, we calculated the estimated marginal mean (EMM) for each randomized group. The T-scores were measured via the Gaussian (normal) distribution, with a truncated minimum and maximum value identified to aid model convergence.
Bayesian statistical inference was used to quantify the probability that model effects exist, given the observed data and weakly informative priors (eg, b ∼ N(µ = 0, σ = 10)). Assumptions of Bayesian inference were evaluated using scale convergence factors (“rhat”), effective sample size, and posterior predictive checking. These assumptions were satisfied. Model inferences relied on the posterior distribution for the regression coefficient for group. The median of the posterior distribution was taken as the most likely point estimate, and a credible interval was derived as the lower and upper limits including 95% of the posterior distribution. The posterior probability (PP) was taken as the proportion of values that were greater or less than the null effect value (eg, b = 0). The literature describes different thresholds of PP values: anecdotal (PP 50% to 74%), moderate (PP 75% to 90%), strong (PP 91% to 96%), very strong (97% to 99%), and extreme (>99%).43⇓–45 The current analyses stipulated that PP ≥ 75% provides a minimum threshold of evidence in favor of the alternative hypothesis to suggest that an effect of group is supported. The protocol was approved by the Committee for the Protection of Human Subjects at the University of Texas Health Science Center at Houston.
Results
We assessed 938 individuals for eligibility (Figure 1). Of these, 11 were excluded due to immediate risk. Participants were randomized to immediate referral (n = 80) or a waitlist (n = 80). Enrollment stopped when the target sample size was reached. While all were referred, only 18 (22.5%) immediate referral and 8 (10.0%) waitlist participants ultimately received services from the MLP. Twelve participants (9 immediate referral, 3 waitlist) withdrew from the study because they no longer needed assistance or did not want to complete the assessments. Follow-up assessment completion (the percentage of participants completing the assessments) ranged from 71.3% to 77.5% in the immediate referral group and from 63.8% to 76.3% for the waitlist. One individual (immediate referral) requested for data to be removed; thus, 159 individuals were analyzed.
CONSORT flow diagram. Abbreviation: MLP, medical-legal partnership.
Baseline Characteristics
Characteristics between groups are reported in Table 1. Individuals averaged 1.3 ED, 0.5 urgent care, and 0.4 hospital visits over the 6-month period before enrollment. Except for the number of participants noting they had 2 or more hospital stays over the past year, the responses to the screening instrument were similar between the 2 groups (Table 2). On average, participants reported having nearly 7 of 25 possible needs. The mean baseline scores were 20.1 for PSS, 24.6 for CES-D, and 9.6 for GAD-7. Compared with the general population in the United States, participants had worse sleep (57.4), pain (61.6), depression (56.7), anxiety (58.9), and fatigue (57.7) and greater difficulty with physical functioning (37.4) and social roles and activities (46.7).
Responses to the Health-Harming Legal Needs Screening Instrument by Total Sample and Randomized Condition
Primary Analysis (6 Months)
At 6 months, immediate referral was associated with lower PSS scores (EMMIMMEDIATE = 18.8, EMMWAITLIST = 19.9; PP = 74.8%) and higher GAD-7 scores (EMMIMMEDIATE = 10.3, EMMWAITLIST = 6.7; PP 89.8%). CES-D scores were not different between the 2 groups (PP = 52.7%). With respect to the PROMIS T-scores, immediate referral was related to worse pain (b = 7.05; PP = 86.9%), social function (b = −6.26; PP = 74.9%), fatigue (b = 8.20; PP = 82.5%), anxiety (b = 6.15; PP = 85.3%), sleep disturbance (b = 4.06; PP = 77.8%), and depression (b = 4.55; PP = 79.3%). Differences across the PROMIS T-scores were not supported for physical function (b = −2.31; PP = 72.7%). Those in the immediate referral group demonstrated a 21% lower rate of ED visits (RR = 0.79; PP = 79.7%), no difference in the rate of urgent care visits (RR = 0.86; PP = 62.0%), and a 75.6% higher rate of hospital visits (RR = 1.76; PP = 88.5%).
Per-Protocol Analysis (6 Months)
Per-protocol analyses evaluated each outcome with respect to receipt of MLP services, comparing receipt of MLP services (REC; n = 26) to nonreceipt (NON; n = 115). The total does not sum to 160 because not all participants completed the client release form (n = 113 completing), and we could not determine the outcome for 18 individuals. Those receiving services were slightly younger age (51.2 vs 53.1 years) and had higher baseline CES-D scores (29.6 vs 22.7). At 6 months, the group that received services demonstrated greater total scores for stress (EMMREC = 23.4, EMMNON = 18.1; PP = 99.2%), depression (EMMREC = 25.3, EMMNON = 15.6; PP = 98.9%), and anxiety (EMMREC = 15.5, EMMNON = 7.5; PP = 98.0%). The REC group demonstrated worse T-scores across all the PROMIS measures: pain (b = 15.7; PP = 97.3%), social function (b = −25.2, PP = 98.6%), physical function (b = −11.3; PP = 99.4%), fatigue (b = 19.3; PP = 95%), anxiety (b = 5.6; PP = 78.3%), sleep disturbance (b = 14.1; PP = 98.7%), and depression (b = 11.4; PP = 96.7%). These groups were not different with respect to health care use (ED visits: RR = 0.93; PP = 58.8%; urgent care visits: RR = 1.18; PP = 63.1%; and hospital visits: RR = 0.96; PP = 53.5%).
Legal Services Provided
Overall, the MLP provided 36 legal services to 30 unique individuals, which includes 4 individuals who were excluded from randomization due to immediate and significant HHLNs (Table 3). Nearly 30% of these benefits encompassed drafting wills, resolving probate issues, and establishing guardianship. In addition, the legal team helped clients to secure child support, seek housing repairs, fight evictions, address employment issues, and obtain Social Security. To address these issues, they provided advice and counsel and wrote letters on behalf of clients.
Legal Benefits Provided by the Medical Legal Partnership
Discussion
Participants in the immediate referral group exhibited lower stress and a 21% lower rate of ED visits but also higher anxiety and a 76% higher rate of hospital visits. Although these results overlap with previous RCTs, unlike them, our findings were not uniformly positive, which, we suspect, is the result of differences in the interventions. First, the previous RCTs focused on pediatric populations; thus, caregivers were the recipients of the services and may have had greater motivation to create stable environments for their children. Second, the prior interventions included additional team members who bridged the medical and social care systems. Of note, the per-protocol analysis demonstrated that receiving legal services was associated with worse outcomes. We hypothesize that participants may have realized that the MLP was unable to resolve their HHLNs causing distress or that new problems may have emerged after HHLNs were addressed. A third possibility is that patients do not want to be connected to legal services through primary care and that these referrals ultimately lead to greater net harms. These will need to be tested in future studies and will be explored during an accompanying qualitative study.
Of the 160 participants referred, only 16% (n = 26), at minimum, received services, highlighting the challenges of integration. This figure mirrors the Accountable Health Communities evaluation, which found that 14% had their health-related social needs resolved.46 In a National Academies of Sciences, Engineering, and Medicine (NASEM) report on primary care and public health, the authors defined integration between the 2 as the linkage of programs to improve population health.8 Our findings suggest that integration is elusive and potentially critical. Using the NASEM nomenclature, our project was a collaboration, because we worked together to plan and execute the program.8 This included joint, biweekly meetings and providing the legal services organization with clinical space. Unfortunately, connecting with patients proved to be difficult, possibly because of a lack of trust and unstable housing, phone, Internet, and transportation. Our experience calls into question the effectiveness of programs that refer to community service providers without integration. Future efforts should aim for and test the effectiveness of partnerships, or “2 entities working so closely together that there is no separation from the end user’s perspective.”8 This would require the legal services organization to be colocated at the clinic, with both sharing as much information as legally allowable when authorized by the patient.
Several limitations should be considered. First, our study randomized referrals rather than the receipt of services. We chose this design to isolate the effect of actions taken by clinics. Future studies should consider randomizing participants after legal services organizations accept them. Second, our results are affected by the lack of a validated screening tool for HHLNs, a problem that plagues social needs screening broadly.47 Developing a screening instrument for HHLNs with acceptable psychometric properties is a critical need. Without such an instrument, efforts to identify individuals with HHLNs may result in high rates of false positives and false negatives. Third, for ethical reasons, we excluded individuals with immediate, serious risks, though their inclusion may have changed our results. Fourth, we were unable to determine the legal outcomes for the nearly 30% that did not provide data-sharing authorization. Finally, we experienced workflow disruptions. Between November 2019 and February 2020, personnel changes at the legal services organization resulted in a coverage gap. The spring of 2020 coincided with the emergence of the COVID-19 pandemic. We highlight these because they reflect real-world challenges that may affect implementation in other settings.
Conclusions
The results of this RCT are mixed, as the MLP was associated with improvements in some outcomes (stress and ED visits) and worsening outcomes in other areas (anxiety and hospital visits). Despite these findings, some received important benefits, including assistance with housing, wills, Social Security, and food stamps. To achieve greater net benefit, practices should strive for tighter integration.
Acknowledgments
We thank our colleagues at Lone Star Legal Aid for their role in providing legal services, Christian Pineda for providing data regarding legal services, Sandra Stansberry for extracting relevant data from the electronic health record, and our colleagues at UT Physicians for their assistance with data collection and referrals.
Appendix.Study Screening Instrument–CONSORT Checklist
Notes
This article was externally peer reviewed.
Funding: The study was supported by a grant from the Texas Medical Center Foundation (WL, principal investigator). The funding agency had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Conflict of interest: None.
Prior presentation: North American Primary Care Research Group. November 20-24, 2020. Virtual.
To see this article online, please go to: http://jabfm.org/content/36/3/414.full.
- Received for publication October 11, 2022.
- Revision received December 8, 2022.
- Accepted for publication December 9, 2022.
