Social determinants of health (SDOH) are the conditions in which individuals live, grow, work and age that shape their health. These conditions include, but are not limited to, education, employment, income, and housing [1, 2]. Stratification based on differences in these conditions leads to unfair and avoidable differences in health, known as health inequities [3–5]. For example, populations that are disadvantaged based on race, ethnicity, immigration, or education have been demonstrated to have higher mortality [1, 6, 7]. Thus, the ability of a healthcare system to identify and address health inequities depends on the availability and accessibility of SDOH data. At a systems level, SDOH data are important not only for health equity but also for evaluations of health systems performance, through risk adjustment. For example, in-hospital mortality is a widely used indicator of health system performance and for benchmarking but is affected by patient-level SDOH such as poverty [8–10]. Not adequately accounting for these factors may result in inaccurate assessments of hospital performance and quality [10]. Therefore, there has been increasing attention to adequate collection and documentation of patient-level SDOH in hospitals, particularly within electronic medical records (EMR), which are longitudinal records of clinical encounters [11–13].

In the EMR, SDOH can exist within structured fields (i.e., where users select the appropriate value or characteristic from a list of options), or in unstructured textual form, such as within clinical notes [14–17]. Despite the potential for an EMR to be a comprehensive repository of patient-level SDOH, there appears to be a significant underreporting of many SDOH in hospitals. For example, we conducted a recent study to develop Natural Language Processing (NLP) methods to extract certain SDOH data, such as education, presence of a language barrier, and living alone and employment status, from an inpatient EMR [18]. We found that within structured fields, while marital status was nearly universally documented, there were zero entries for occupation. Such variation across different SDOH variables is consistent with the literature, primarily from the United States, suggesting that while core demographics such as legal sex are generally highly complete in EMRs, other data such as gender identity, social support and housing status are under-documented [19–22]. In addition, SDOH tend to be more commonly captured in notes in the EMR (as unstructured data, requiring extraction such as through NLP) rather than as structured data [21–24]. Current hospital-based studies on SDOH documentation, however, tend to focus on a narrow set of SDOH variables and elements, and/or a specific disease cohorts and populations, limiting their generalizability [23–25].

What remains unknown in the current evidence base is how reporting rates of SDOH data vary across a wide range of variables especially within the hospital context, and the extent to which patient factors influence these rates of capture. This information is important to ascertain for several reasons. First, knowledge of which SDOH variables are poorly documented in EMR is a necessary antecedent to improving the documentation of SDOH. Second, as the use of NLP grows in extracting SDOH from unstructured notes from EMR, efforts to improve these technologies should focus on the variables that are highly prevalent (i.e., there is little value in refining the extraction of data when the data do not exist). Lastly, the exploration of factors that influence SDOH documentation can inform current and future interventions and quality improvement initiatives to address barriers related to SDOH data collection and documentation. We therefore undertook a cross-sectional study of 10,659 inpatients in Calgary, Alberta, Canada between 2017 and 2022 to determine the prevalence of documentation of nine SDOH variables (immigrant status, refugee status, ethnicity, marital status, employment status, presence of financial strain, homelessness, living alone status, and gender identity), and the influence of patient and hospitalisation factors that influence these documentation rates.

The study cohort consisted of a total of 10,659 patients, of whom 52.2% were females (see Table 1 for sample characteristics). The median age of the study cohort was 62 years, and the median length of stay was 4 days. Most patients (56.4%) had no Charlson comorbidities, while 15.3% had 2 or more Charlson comorbidities. The study cohort was stratified into two physician specialties, surgical and non-surgical, and 65.8% of the patients were from the non-surgical group.

Table 2 shows the documentation rates of the different SDOH variables. Marital status was the most frequently documented (87.44%) SDOH variable, followed by homelessness status (81.93%) and living alone status (71.83%). Amongst the least captured SDOH variables were refugee status (0.34%), immigrant status (5.62%) and ethnicity (7.14%). In stratified analysis (Table 2), patients who had longer hospital stays (≥ 8 days) were observed to have higher documentation rates of each SDOH variable compared to those who had shorter hospital stays. Similarly, older patients (≥76 years) had higher documentation rates of all SDOH variables, with the exception of employment status. Patients who had multiple comorbidities (≥2) were also found to have higher documentation rates of SDOH variables compared to those with only 1 or no Charlson comorbidities (Table 2).

Table 3 presents the adjusted odds ratio demonstrating associations between documentation of SDOH variables and the patient and hospitalisation variables. Odds of documentation of SDOH variables amongst surgical cases were lower than non-surgical cases. For instance, surgical cases had decreased odds of documentation of ethnicity (aOR 0.39, 95% CI: 0.31-0.47), immigrant status (aOR 0.33, 95% CI: 0.26-0.42) and financial strain (aOR 0.43, 95% CI: 0.37-0.51). Odds of documentation of most SDOH variables were observed to increase with an increase in the number of Charlson comorbidities. Similarly, the odds of documentation increased with an increase in the length of hospital stay. For example, the odds of documenting ethnicity increased by 1% with an increase of one day in the LOS (aOR 1.01, 95% CI: 1.01-1.01), whereas the odds increased by 15% for each additional Charlson comorbidity recorded (aOR 1.15, 95% CI: 1.05-1.25). These findings were statistically significant (p<0.001).

Having recorded female sex in DAD was associated with lower odds of documentation of employment status (aOR 0.48, 95% CI: 0.44-0.53) and financial strain (aOR 0.73, 95% CI: 0.65-0.83). These findings were statistically significant (p<0.001). These logistic regression results are congruent with the higher rates of documentation of employment status in males compared to females, as reported in Table 2. Gender identity was more likely to be documented in patients recorded as female sex compared to patients recorded as male sex (aOR 1.18, 95% CI: 1.09-1.29). The association of female sex with other SDOH variables was not found to be statistically significant. Age was not observed to be associated with documentation rates of most SDOH variables, with the exception of employment status. Marital status, as reported in Table 2, was well documented and not impacted by any of the patient variables – possibly due to the existence of a known structured field in the EMR for this data element [18]. Refugee status, which had a very low documentation rate, did not have any significant association with any of the patient/hospitalisation variables studied.

Finally, Table 4 shows how the documentation rates have changed over time. There was a gradual increase in documentation rates over time for most variables, with the largest occurring in living alone and employment status.

In this study of 10,659 inpatients from four acute care facilities in Calgary, Alberta, Canada, we assessed rates of documentation of nine SDOH variables. We found substantial variability in the rates of documentation across different SDOH, with marital status as the most frequently documented (with 87.44% of all charts having this recorded), followed by homelessness (81.93%) and living alone status (71.83%). Refugee status, immigration status, ethnicity, and financial strain were amongst the least documented SDOH variables. Complex patients with higher numbers of comorbidities and longer hospital stays had significantly higher odds of documentation of SDOH, whereas admission to surgical services had lower rates of SDOH documentation. We also observed an increase in the documentation rate of most SDOH variables over the study period, with the largest changes seen in employment status going from a documentation rate of 25.35% in 2017 to 40.62% in 2022 and living alone going from 67.25% to 78.37% in that period. This finding may reflect the growing recognition of the importance of SDOH collection in the healthcare system.

Our findings on the variation of documentation rates across different SDOH are consistent with existing evidence, which is based almost entirely within the United States. These studies show that identity and demographic fields – such as race and legal sex – are nearly universally complete in EMR because they are mandated at registration, billing, or for reporting purposes [19–21, 33]. Unlike the United States, however, Canada does not mandate the capture of race or ethnicity. The result of this is, as expected, lower documentation of ethnicity – we found that only 7.14% of charts contain this information. Existing literature also suggests that socioeconomic and social risk elements are severely under-documented in the inpatient setting; they are virtually absent within structured fields [17, 21], but even when considering notes and unstructured text, they remain significantly under-documented [21–24, 34]. For example, in one study, the use of text mining techniques detected the presence of unstructured variables such as housing, financial strain, and social isolation in only 1-3% of nearly 1.2 million charts [21]. Our study similarly shows the near-complete absence of certain SDOH variables (such as refugee status); however, our estimates of documentation for financial strain, living alone status, and homelessness status are higher than those reported in the literature. Therefore, while it is true that documentation rates across different SDOH variables do vary widely, which variables are highly documented versus not and the absolute rates of their completion depend considerably on regional and national contexts.

There is a substantial body of evidence that explores barriers to the collection of SDOH data. At the patient level, while patients are generally open and receptive to the collection of these data, there remain concerns about confidentiality, perceived benefit, and potential stigmatisation [35–38]. At the provider level, barriers to SDOH data collection include workload and time demands, perceived inability to act on such data, comfort with asking patients about SDOH, and differences in opinion about who is responsible for SDOH collection [36, 37, 39–42]. With increasing attention to, and calls for, universal collection of SDOH data in hospital [12, 13], there is a need not only to understand these perspectives, experiences, and individual-level barriers (to inform considerations of how SDOH data should be collected) but also systematic gaps in the current collection and documentation of SDOH data (to inform targeted areas of focus for future intervention). Our study provides this important exploratory information. For example, we found that patients who have fewer comorbidities and shorter hospital stays have SDOH less commonly recorded in their EMR charts. Similarly, despite the importance of SDOH data in predicting surgical and post-surgical outcomes [43–45], patients admitted to surgical services have lower odds of having their SDOH documented. These populations are therefore important targets for interventions aimed at improving inpatient SDOH collection. These findings also shed light on the subpopulations that might be systematically “missing” when using SDOH and social barrier data extracted from EMR to inform program development and policy planning to identify and address health inequity.

Despite the presence of these barriers, there remains interest in, and calls for, the widespread and systematic collection of SDOH data across entire patient populations. This interest stems from the recognition that SDOH data could be used not only to personalise and improve patient-centred care at the individual level, but is also vital at a systems level for health systems evaluation, program planning, and identification and mitigation of health inequities. A 2022 systematic review examined the integration of SDOH screening in EMR and the effects of interventions and quality improvement initiatives targeting SDOH. This review found these SDOH interventions consistently demonstrated improved process outcomes as well as improved access to resources and resolution of patient-identified concerns [46]. However, studies reporting outcomes on healthcare utilisation and cost outcomes were both limited in number and reported mixed results [46]. More recently, the Centers for Medicaid and Medicare Services in the United States implemented and tested the Accountable Health Communities Model, with 28 participating organisations [47]. Beneficiaries residing in the community were comprehensively screened for social needs, referred to community services, with a subset being provided with navigation services, while simultaneously strengthening the partnerships between clinical partners and community organisations. This intervention resulted in reduced readmissions, emergency department visits, and ambulatory care visits, generating an estimated $200 million in net savings [47]. Ultimately, the scale of health inequity attributable to unmet social needs, combined with this emerging evidence suggests that systematic identification and documentation of SDOH remains a necessary foundation and prerequisite for the development and coordination of clinical, community, and systems-level interventions to produce downstream benefits.

The strengths of our study include an extensive, large-scale, rigorously implemented chart review of over 10,000 charts to obtain SDOH documentation rates. We therefore provide a comprehensive view of the prevalence of SDOH documentation in the EMR, not limited to structured fields (which tend to have significant under-documentation of SDOH) and not limited by the performance of the approaches used (such as with natural language processing) to extract SDOH from unstructured notes. This makes the study unique and notable in the field. Another strength is the broad scope, across multiple centres and admission services, without limiting to any specific disease cohort. Despite these strengths, we acknowledge certain limitations of this study. First, while we included a broad population across multiple centres (i.e., all four hospitals in the city of Calgary), all centres were located in one city in Canada, and all used the same single EMR system. This may limit generalizability to other jurisdictions, which may have different mandates, workflows, processes, and EMR platforms that may influence whether, which, and how SDOH are collected. Further, since we only included records from inpatient stays at acute care hospitals, generalizability of our findings is limited to other settings such as urgent care facilities and outpatient and/or primary care clinics. Second, our documentation rate estimates do not differentiate between structured and unstructured data. There have been numerous studies on this topic, consistently showing higher documentation of SDOH in unstructured notes compared to structured fields, so this line of inquiry was not further pursued; such information would have been helpful in data interpretation (e.g., whether the existence of structured fields like marital status was associated with improved documentation of certain SDOH variables). Third, the SDOH variables of interest in this study were selected based on study team discussion and consensus and informed by our team’s prior work [18]. While established SDOH frameworks, such as those proposed by the World Health Organization [4], identify a broader range of structural and social determinants than those examined in this study, the variables selected represent individual-level factors commonly documented in clinical records and relevant to the inpatient setting. The absence of a formal SDOH framework to guide variable selection may have resulted in the omission of relevant SDOH variables. Fourth, the low documentation rates of certain SDOH variables, such as refugee status, result in few documented cases especially within certain patient subgroups (e.g., patients under 46 years, or surgical patients). Effect estimates and significance tests for these variables should be interpreted with caution. Lastly, separate logistic regression models were run for each SDOH variable to assess the associations between patient factors and SDOH documentation rates. While this approach is aligned with our research objective (i.e. to assess documentation rates across different SDOH variables), which acknowledges that documentation rates and the factors that influence them are likely to vary across different SDOH variables, we also recognise that this approach would not capture co-documentation patterns or correlations between SDOH variables. In particular, failing to account for co-documentation patterns means that we cannot determine whether certain patient factors are associated with systematically lower documentation across multiple SDOH domains simultaneously, which may have different clinical and health equity implications. Approaches that jointly model multiple correlated outcomes or identify subgroups of patients based on overall documentation could be addressed in future work.

There is a need for comprehensive patient-level SDOH data in hospital-based EMR. These are essential for program development, health systems evaluations, and improving equity of outcomes and delivery of care. Despite their importance, we found that there is wide variation in documentation across different SDOH variables. Unless this documentation can be improved, the intensive efforts aimed at developing and improving technologies to extract certain SDOH data from EMR notes may be of limited utility [16, 48, 49]. Our findings also inform future interventions and initiatives to improve SDOH documentation by highlighting systematic gaps, such as severely under-documented SDOH variables and populations that are less likely to have SDOH documented, in current practice. Addressing these gaps and strengthening SDOH collection and documentation is essential to guide equitable care and system improvement.