Results and discussion

Of 480 systematic reviews identified, 8 duplicates were removed, and 125 abstracts were selected for full-text review. These 125 systematic reviews were assessed, and 39 reviews (9%) were included in the final analysis (Figure 1). Our inclusion percentage is in line with high quality systematic reviews (2–8%). ^{Reference Wang, Nayfeh, Tetzlaff, O’Blenis and Murad7} According to the SANRA scale, the quality of this narrative review was very high (score:10, Supplement 2). Below are summarized responses to the questions that guided our search strategy:

1. I. What is the association of race, ethnicity, SDOH with quality measures? (n = 23) A total of 39 articles included data regarding the association of race, ethnicity, and SDOH with six specific inpatient quality measures (Table 4). We did not identify articles that addressed the remaining 23 inpatient quality measures (Table 2).

   • 30-day readmission: Six reviews evaluated the association of race, ethnicity, or SDOH with 30-day readmission rates. Age (n = 4), sex (n = 4), race (n = 3), and income (n = 3) were most evaluated. Of these reviews, the majority demonstrated that patients with increasing age, male sex, and “non-white race” (as reported in systematic review) had increased risk of 30-day readmissions. One review evaluating inpatients admitted with cardiac conditions concluded that female sex was associated with increased risk of 30-day readmission and that female inpatients were less likely to receive a cardiac intervention compared to male patients. ^{Reference Hoang-Kim, Parpia and Freitas8} Some reviews concluded that public insurance (n = 2), obesity (n = 1), and unemployment status (n = 1) were associated with increased risk of 30-day readmission. Additionally, one review concluded that higher income was associated with a decreased risk of any hospital readmission (ie, not limited to all-cause 30-day readmission) while lower levels of social integration, including participation in a broad range of social relationships as measured by the Social Network Index, was associated with an increased risk of any type of hospital readmission. ^{Reference Desai, Stauffer, Feringa and Schreiner9}

   • All-cause mortality: Twenty-three reviews evaluated the association of race, ethnicity, or SDOH with all-cause mortality. The most evaluated measures were sex (n = 6), race (n = 5), exercise (n = 5), and nutrition (n = 3). ^{Reference Cohen, Doyle, Skoner, Rabin and Gwaltney10} Among these reviews, the majority demonstrated male sex, “non-white race,” and frailty were associated with increased risk of all-cause mortality, while increased physical activity and improved nutrition were associated with decreased risk. A smaller number of reviews concluded that poverty, higher levels of pollution exposure (particulate matter), lower socioeconomic status, or immigrant status were associated with increased risk. A single review concluded that age >75 years was associated with increased risk, ^{Reference Damayanthi, Prabani and Weerasekara11} while a different individual review concluded that higher level of employment was associated with a decreased risk. ^{Reference Tao, Zeng, Liu and Fu12}

   • Falls: Four reviews evaluated the association of race, ethnicity, or SDOH with inpatient falls. All four evaluated the association of exercise with falls; the majority concluded that increased exercise and physical activity were associated with decreased risk of falls, including both inpatient falls and falls in all settings. A single review concluded that use of alcohol or other drug use was associated with increased risk of falls. ^{Reference Lau, Ang and Kim13} Similarly, a different review concluded that language barriers and/or a communication disability were associated with increased risk of falls among neurology patients. ^{Reference Hemsley, Steel and Worrall14} One review was inconclusive about the association of nutrition and vitamin D levels with the risk of falls. ^{Reference Guirguis-Blake, Michael, Perdue, Coppola and Beil15}

   • Length of stay: Three reviews evaluated the association of race, ethnicity, or SDOH with hospital length of stay. Two reviews concluded that “non-white race” was associated with a longer length of stay. ^{Reference Mendoza, Pangal and Cardinal16,Reference Mo, Ikwuezunma and Mun17} Similarly, a single review concluded that lower levels of social integration were associated with longer length of stay. ^{Reference Valtorta, Moore, Barron, Stow and Hanratty18}

   • Surgical Site Infections (SSI): Two reviews evaluated the association of race, ethnicity, or SDOH with SSIs. One review concluded that “non-white race” was associated with increased risk of SSI. ^{Reference Mo, Ikwuezunma and Mun17} A separate review concluded that obesity was associated with SSI. ^{Reference Partha Sarathi, Mowforth and Sinha19}

   • Venous thromboembolism (VTE): One review evaluated the association of race, ethnicity, or SDOH with risk of VTE. ^{Reference Barrett, Whitehouse, Blom and Kunutsor20} This single review concluded that increasing age, increasing BMI, female sex, and African American race were associated with increased risk of VTE.

   Figure 1.

   Flow diagram of the narrative review process.

   
   Table 4.

   Narrative review of existing systematic reviews describing associations between race, ethnicity, social determinants of health and quality measures, 2010–2022, N = 39

   

   * Reported in table as defined/reported in systematic review

2. II. How are race, ethnicity, and SDOH measures defined in the context of healthcare quality? (n = 10)

Among the 39 systematic reviews, 10 reviews included definitions for race, ethnicity or SDOH and examined their association with inpatient quality measures, specifically all-cause mortality and alcohol-attributable mortality. In the reviews that clearly defined SDOH measures, there was a heavy reliance on available metrics of income, insurance, poverty indices, education, housing, and crime to define socioeconomic statuses. ^{Reference Probst, Roerecke, Behrendt and Rehm21–Reference Cusimano, Pshonyak, Lee and Ilie27} Specifically, reviews focused solely on single measures when it related to frailty or migration. However, reviews that used socioeconomic status measures typically included multiple measures (eg, income, residence, and insurance type, etc.). To facilitate comparisons, these reviews restricted their included studies to either a single SDOH measure or to their predetermined SDOH definition, in particular for socioeconomic status.

Race and ethnicity categorizations were either not clearly defined or variably defined often relying on U.S. Census Bureau definitions or other similar agencies. ^{Reference Cusimano, Pshonyak, Lee and Ilie27} For example, in an included review of “indigenous” adults with rheumatic diseases, “indigenous” was defined as Canadian First Nations, Native Americans and Australian Aborigine to account for populations across Canada, Australia, and the United States. ^{Reference Hurd and Barnabe26} In another study on infant mortality among immigrants, country of birth was used as a proxy for race while ethnicity was reported as an “undefined term” representing national origin or citizenship. ^{Reference Gissler, Alexander and MacFarlane25} Also of note, in a number of reviews, SDOH factors with subjective definitions were used. For example, Valtorta et al. evaluated the social relationships of older adults by defining categories of isolation and loneliness based upon availability of perceived support. ^{Reference Valtorta, Moore, Barron, Stow and Hanratty18}

1. III. How complete and accurate is documentation of race and ethnicity and SDOH in observational health databases, including EHRs? (n = 3)

While basic demographic data (eg, age, race, ethnicity, insurance type) are often included as part of routine EHRs to facilitate medical billing and/or to fulfill federal reporting requirements, only three reviews assessed data completeness, while no reviews assessed data quality. One study evaluating patients older than 15 years of age presenting to the hospital with falls noted a range of 1–88% of patient records had missing documentation for alcohol use or alcohol consumption. ^{Reference Lau, Ang and Kim13} Despite the lack of “best practices” to address data completeness and accuracy, one review by Cook, et al. sought to assess patient-level data in EHRs. ^{Reference Cook, Sachs and Weiskopf28} Their review searched for race, ethnicity, language preference, health insurance status, country of origin, occupation, socioeconomic status, education level, environmental health determinants (eg, proximity to food sources, walkability, exposure to toxins), and geocoded patient address data as it was documented in the EHR. Their analyses revealed higher rates of race and ethnicity misclassification and poor data quality for some racial groups. Specifically, assumptions were made by data recorders around the ethnicity of patients with “Spanish sounding” last names who were, in fact, neither Hispanic nor Latino, as well as low data accuracy for patients identified as having “Hispanic or Latino” ethnicity in general. ^{Reference Cook, Sachs and Weiskopf28}

Among the included systematic reviews, few included any documented measures of disability, information on socioeconomic status, language, interpreter use, literacy, or numeracy data. Overall, there appears to be minimal consensus on which SDOH measures to collect besides those required for federal reporting and/or billing purposes. To account for variation in available EHR documentation, many reviews included external data sources such as census data or aggregate zip code information which are approaches with several known limitations. These external data sources lose key patient-level data in the aggregation process, may solely focus on income or education as proxy measures of SDOH, and may not adequately capture intersectionality among SDOH measures. ^{Reference Mahmoudi, Kamdar, Kim, Gonzales, Singh and Waljee29}

1. IV. How often are patients admitted to the hospital intentionally screened for SDOH measures? (n = 0) No systematic reviews addressed this question. An additional targeted search of the literature for research on screening for SDOH revealed the following information:

2. Screening tools: A wide variety of published tools exist for potential use in hospital screening, each with distinct lengths, modes of administration and areas of focus. ^{Reference Davis, Rodger and Pinto30} For example, the Protocol for Responding to and Assessing Patient’s Risks, Assets, and Experiences is a tool developed for and by community health centers, is available in over 30 languages (pending validation in these languages), and is increasingly used across healthcare settings. ³¹ The Accountable Health Communities Health-Related Social Needs screening tool was developed with expert input and tested by the Centers for Medicare & Medicaid Services for use in a range of clinical delivery settings. ^{Reference Billioux, Verlander, Anthony and Alley32} The Institute of Medicine has identified SDOH domains for the meaningful use of EHRs, and the Office of the National Coordinator for Health Information Technology (ONC) has established interoperability standards to facilitate the sharing of SDOH data elements. ³³ With this in place, there is potential for wider implementation of SDOH screening programs facilitated through EHRs. However, there are no agreed-upon national standards for which data elements should be routinely collected or how these data should be applied to achieve meaningful use. A recent scoping review revealed the most frequently assessed SDOHs in studies limited to inpatient general internal medicine wards are food security, finances, housing and transportation. ^{Reference Davis, Rodger and Pinto30} Similarly, there was a lack of current standard on the best methods of administering these screening tools, based on preferred language spoken, literacy, and/or education level; existing reviews utilized data collected by self-report with a written questionnaire (which excludes other language speakers, vision-impaired, and preliterate patients), physician administered, or administration by dedicated staff (with no documentation of interpreter use for other language speakers). Previous reviews have demonstrated that patient responses are influenced by fears around the implications of this data collection. ^{Reference Pinto, Glattstein-Young, Mohamed, Bloch, Leung and Glazier34}

3. Implementation of screening and interventions: Most of the available literature consists of implementation reviews assessing the introduction of a SDOH screening tool to a specific clinical site. The majority of these reviews were performed in the outpatient setting and showed that implementation of these screenings was feasible with large numbers of unmet patient needs identified. ^{Reference Bechtel, Jones, Kue and Ford35} Ability to address identified needs, however, varied greatly by the resources available in the given clinical context. There are limited number of systematic reviews looking at universal implementation of adult inpatient screening for SDOH. One systematic review identified only 8 studies, of which 5 interviewed patients directly, all in English only. The largest of these reviews included approximately 1400 patients, a small sample size. ^{Reference Davis, Rodger and Pinto30} Healthcare system resources needed to implement systematic screening and SDOH barrier interventions were not addressed in these reviews.

4. V. How often are patients identified as having one or more SDOH risk factor?(n = 0)

Our search strategy failed to identify any published systematic reviews to address the prevalence of SDOH risk factors in hospitalized patients and sub-populations. Given that universal inpatient SDOH screening has not been implemented, accurate estimates of risk factor identification are not available. Based on articles identified to address screening implementation (see Question IV, above), estimates were available from small reviews. In these small reviews, the rates of identified SDOH risk factors were highly variable depending on the screening tool used and the underlying population assessed. Two of the largest reviews are summarized, below, demonstrating this variability.

Meyer et al. looked at implementation of universal screening for SDOH at New York-Presbyterian Hospital and associated primary care sites. ^{Reference Meyer, Lerner, Phillips and Zumwalt36} Over the review period, 13,273 patients in North Manhattan were screened; 82% identified as Hispanic, 14% identified as Black/African American and the average household income was $24,000. In this population, 27% screened positive for food insecurity, 25% screened positive for housing insecurity, 12% screened positive for transportation needs, 8% screened positive for utility needs and 1% screened positive for safety needs. Smith et al. analyzed 1,427 patients admitted to the general medicine service at an academic health center in Toronto, Canada. ^{Reference Smith, Kuluski and Costa37} Their review population was less racially diverse (69% white, 11% Asian, 5% Black, 14% “other”). Income data were limited (nearly 50% of the population selected “prefer not to answer”; 14% reported an income below $20,000 USD; 12% reported challenges with English language that inhibited their care; 54% reported having at least one disability. Food and housing security, transportation needs, and safety needs were not addressed in this review.

1. VI. How often are race, ethnicity and SDOH measures reported when describing specific hospital quality measures?

The most evaluated demographic and SDOH measures were sex (n = 9), income and/or employment status (n = 9), age (n = 6), race (n = 6), and education (n = 5). We found very limited assessment of other SDOH measures such as economic stability, neighborhood, language spoken, and health systems access in the included systematic reviews. SDOH measures were most assessed related to the hospital quality measures of all-cause mortality and readmission rates; few addressed increased risk for falls, venous thromboembolism, or length of stay; no systematic reviews assessed medication use errors or healthcare-associated infections (HAIs) besides SSIs (Table 4).