Social Determinants of Health Coding Among Hospital Visits Made by Adults With Cancer

Abstract

PURPOSE

Social determinants of health (SDOH) affect clinical outcomes among patients diagnosed with cancer. In 2015, International Classification of Diseases-10 diagnosis codes starting with Z prefix, or Z-codes, were introduced to capture SDOH complexity. We examined Z-code uptake for patients treated by an academic medical center and county safety-net health system and their comprehensive hospital visits from a regional health information exchange.

METHODS

We identified adults age 18 years or older from two tumor registries, diagnosed from January 1, 2015, to December 31, 2023, and identifiably linked them to their longitudinal, comprehensive hospital visits. We characterized Z-code uptake trends as patients visited over 80 hospitals, for all payers including the uninsured.

RESULTS

We identified 57,598 patients with 181,684 associated ED and inpatient hospital discharges across more than 80 regional hospitals. Approximately 4.7% of patients had at least one Z-code entered during any hospital visit, and Z-codes were present in just 2.8% of all hospital discharges. Z-code use rose over time from 1.1% of visits in 2016 to 11.3% by 2023. Of the cohort, 23.9% were treated by the county safety-net system; 9.4% had Medicaid and 28.6% were uninsured/charity care at diagnosis; and 24.4% had advanced-stage disease. The most common Z-code was Z59 (housing and economic circumstances), accounting for 48.6% of all Z-codes. Documentation was nearly three-fold higher at the county safety-net system than at the academic center (8.6% v 2.9%; P < .001).

CONCLUSION

In a population-based sample of adults with cancer and their comprehensive hospital visits, Z-code use increased to over 11% of hospital visits by 2023, and housing and economic problems were most prominent. Future work should explore the validity of this coding at an individual patient level.

INTRODUCTION

Identifying and addressing the unmet social needs of patients diagnosed with cancer is critical to providing holistic cancer care. Social determinants of health (SDOH) capture the nonmedical conditions (social, economic, and environmental) that also affect health outcomes.¹ In October 2015, the International Classification of Diseases (ICD) evolved to include SDOH codes with the Z prefix.² Although ICD codes with the Z prefix may denote various health conditions (such as pregnancy), SDOH ICD codes were created from Z55-Z65, referred to as Z-codes hereafter.³ These are intended to allow reporting of the complexity of caring for patients with these unmet social needs.⁴

However, use of Z-codes has been limited, largely due to lack of regulatory requirement and financial compensation for their use.⁵ For instance, only 0.65% of emergency department visits were noted to have an associated Z-code in 2016, with a gradual increase to 1.17% by 2019.⁶ Among Z-codes entered between 2017 and 2021, the most commonly documented socioeconomic needs were those related to housing.⁷ Eventually, the Centers for Medicare & Medicaid Services (CMS) added SDOH needs screening for all inpatient admissions beginning January 1, 2024.⁸ Specifically, the Screening for Social Drivers of Health and Screen Positive Rate for Social Drivers of Health were integrated into the Hospital Inpatient Quality Reporting (IQR) Program, with voluntary data collection in 2023 and mandatory hospital-level reporting for 2024 onward. Because IQR performance affected a hospital system’s payment, this was one of the first direct financial incentives associated with identification of social needs to hospital reimbursement. In addition, policy changes have introduced reimbursement opportunities for other activities, including risk assessment and care navigation services, including the HCPCS G0136 billing code for conducting SDOH screening assessments.⁹ This was to lay the groundwork for required reporting of inpatient screening rates beginning 2024, with potential for more stringent screening requirements in subsequent years.

Thus far, the bulk of the literature has reported on Z-code uptake among general patient populations. Unaddressed SDOH also negatively affect access to, and outcomes of, cancer treatment.^10-13 There is limited information regarding Z-code usage among adults with cancer. Our goal was to characterize Z-code use and prevalence across comprehensive hospital visits from a regional health information exchange, linked to a combined population-based tumor registry cohort from an academic medical center and county safety-net hospital.

METHODS

Cohort, Setting, and Data Sources

We examined a population-based sample of adults age 18 years and older, with incident diagnoses in two local tumor registries from January 1, 2015, to December 23, 2023. The two local registries included patients from University of Texas Southwestern (UTSW) Medical Center, an academic medical center, and Parkland Health (PH), an academically affiliated, but clinically distinct, county safety-net hospital for Dallas County. PH is virtually the only provider of cancer treatment for the uninsured in Dallas County—a county with one of the highest working-age uninsured rates in the country—via its indigent care program.¹⁴ Both health systems have operated on the Epic electronic medical record (EMR) since before 2015.

Each patient’s record was also identifiably linked to their comprehensive hospital visits through a regional health information exchange, enabling longitudinal tracking across more than 80 hospitals within a 100-mile radius of Dallas, TX (all nonfederal hospitals in the region).¹⁵ This linkage provided administrative data on hospital visits, including diagnoses, for all insurance types, including the uninsured. For patients with multiple, synchronous cancer diagnoses, the highest-stage cancer was selected. For patients with metachronous cancer diagnoses, the first diagnosis was included, and subsequent diagnoses were excluded. Advanced stage was defined as stage IIIB or higher for lung cancer, stage III or higher for pancreatic cancer, and stage IV for all other cancers except for brain cancer.^16,17

Primary Outcome and Variables of Interest

The primary outcome of interest was the presence of a Z-code associated with encounters. We used ICD-10 codes Z55-Z65, which encompass problems related to education, literacy, employment, housing, social environment, upbringing, primary support group, and other psychosocial circumstances. The tumor registry and EMR provided age at diagnosis, sex, race and ethnicity, cancer type, and stage at diagnosis. We recorded primary payer at the time of hospital visit from the health information exchange. Hospital visits included both emergency department visits and inpatient hospitalizations.

Statistical Analysis

We used descriptive statistics to characterize the frequency and trends in Z-code use. We then identified patients with at least one Z-code identified during any hospital encounter. We examined sociodemographic and clinicopathologic associations with Z-code entry. We then analyzed whether the primary cancer-treating hospitals identified SDOH more often than nontreating hospitals. A multivariable logistic regression was used to confirm patient characteristics associated with at least one Z-code entry versus no Z-code entry.

Chi-square tests of significance were used to compare categorical data, and Wilcoxon rank sum tests were used to evaluate continuous data with a significance level of .05. All statistical analyses were performed using R (version 4.4.2).¹⁸ This research was approved by the UTSW Medical Center Institutional Review Board (STU-2023-0256).

RESULTS

As shown in Table 1, we identified 57,598 patients with at least one hospital discharge. In total, 18,017 (31.3%) patients were treated by the county health system. Our cohort included 29,951 (64.1%) White, 13,489 (27.5%) Black, and 1,765 (3.6%) Asian or Pacific Islander patients; 10,385 (21.2%) identified as Hispanic or Latino. The primary payer at time of visit included 14,013 (28.6%) self-pay or charity, 14,943 (30.5%) commercial health insurance, 12,456 (25.4%) Medicare, and 4,617 (9.4%) Medicaid. The most common cancers were breast (15.7%), hematologic (12.6%), and lung (10.4%).

TABLE 1.

Baseline Sociodemographic and Clinicopathologic Characteristics of Patients

Characteristic	Cohort Total (N = 57,598)	Patients With No Z-Code (n = 54,903)	Patients With Z-Code (n = 2,695)
Age at hospital visit, years, mean (SD)	58.3 (14.8)	58.5 (14.9)	55.3 (13.1)
Payer status
Charity care/uninsured	14,013 (28.6)	12,690 (27.4)	1,323 (49.1)
Commercial health plan	14,943 (30.5)	14,615 (31.5)	328 (12.2)
Medicare	12,456 (25.4)	12,032 (25.9)	424 (15.8)
Medicaid	4,617 (9.4)	4,144 (8.9)	473 (17.6)
Other	3,044 (6.2)	2,900 (6.3)	144 (5.3)
Race
White	29,951 (64.1)	28,324 (61.2)	1,567 (58.2)
Black or African American	13,489 (27.5)	12,524 (28.8)	965 (35.1)
Asian or Pacific Islander	1,765 (3.6)	1,713 (3.7)	52 (1.9)
American Indian	131 (0.3)	122 (0.3)	9 (0.3)
Other race	3,754 (7.6)	3,655 (7.9)	99 (3.7)
Hispanic/Latino ethnicity	10,385 (21.2)	9,573 (20.6)	812 (30.2)
Female	24,546 (49.8)	23,403 (50.2)	1,143 (42.4)
Advanced stage at diagnosis	14,027 (24.4)	13,225 (24.1)	802 (29.8)
Cancer type
Brain	1,883 (3.3)	1,826 (3.3)	57 (2.1)
Breast	9,053 (15.7)	8,644 (15.7)	409 (15.2)
Lung	5,966 (10.4)	5,704 (10.4)	262 (9.7)
Leukemia/lymphoma	7,331 (12.8)	7,046 (12.9)	285 (0.5)
Kidney	1,473 (2.6)	1,391 (2.5)	82 (3.0)
Thyroid	2,363 (4.1)	2,278 (4.1)	85 (3.2)
Head and neck	3,553 (6.2)	3,382 (6.2)	171 (6.3)
Prostate	4,226 (7.3)	4,027 (7.3)	199 (7.4)
Pancreas	2,281 (4.0)	2,195 (4.0)	86 (3.2)
Gastrointestinal/colorectal	9,975 (17.3)	9,346 (17.0)	629 (23.3)

NOTE. Cohort demographics of adults diagnosed and treated for incident cancer at an academic medical center and county safety-net health system in North Texas. Columns compare the entire cohort, patients who had a SDOH need coded during a hospital visit after diagnosis, and patients who did not have a SDOH need coded.

Abbreviations: SD, standard deviation; SDOH, social determinants of health.

During the study period, 4.7% (2,695/57,598) of patients had at least one Z-code entered. In total, there were 3,572 Z-codes entered. The most common Z-codes were Z59: housing and economic circumstances (1,735/3,572, 48.6% of all Z-codes) and Z63: problems with family circumstances (482/3,572, 13.2%). The Data Supplement (Table S1, online only) shows the proportion of hospital visits with a Z-code over time. In 2016, there were 1.1% (227/20,460) of hospital visits with a Z-code. By 2022, this percentage increased to 3.2% (710/22,222), reaching 11.3% (2,168/19,180) by 2023. These trends are depicted in Figure 1.

FIG 1.

Time trend of comprehensive regional hospital visits made by adults with cancer who had a social determinant of health Z-codes coded, from 2015 to 2023.

In multivariable logistic regression (Table 2) to identify characteristics associated with Z-code entry, Black race (odds ratio [OR], 1.09 [95% CI, 1.01 to 1.17]) compared with White, being uninsured (OR, 2.76 [95% CI, 2.46 to 3.10]) compared with commercial, treatment in the county system (OR, 1.83 [95% CI, 1.69 to 1.98]) compared with academic medical center, and having a hospital visit in 2023 (OR, 4.3 [95% CI, 3.89 to 4.76]) compared with 2016-2022 were all associated with greater adjusted odds of Z-code entry. In comparison, age above 65 years (OR, 0.67 [95% CI, 0.61 to 0.73]) compared with 18-64 years and Hispanic ethnicity (0.50 [95% CI, 0.46 to 0.55]) were associated with less Z-code entry.

TABLE 2.

Multivariate Logistic Regression Analyzing Patients With at Least One Z-Code Entry Versus None

Variable	P	aOR	Lower 95% CI	Upper 95% CI
Age 65 years and above (ref = 18-64)	<.001	0.67	0.61	0.73
Male gender	.62	0.98	0.92	1.04
Race and ethnicity (ref = White)
Black or African American	.02	1.09	1.01	1.17
Asian or Pacific Islander	<.001	0.42	0.32	0.53
Hispanic	<.001	0.50	0.46	0.55
Other	<.001	0.67	0.54	0.81
Payer status (ref = commercial)
Uninsured (charity care)	<.001	2.76	2.46	3.10
Medicare	<.001	1.61	1.41	1.83
Medicaid	<.001	2.80	2.49	3.15
County hospital system	<.001	1.83	1.69	1.98
Year of hospital visit, 2023 (ref 2016-2022)	<.001	4.30	3.89	4.76
Advanced cancer stage at diagnosis	.71	1.01	0.94	1.09

NOTE. Multivariate regression analysis estimating the relationship between patient and system characteristics associated with greater likelihood of coding of social determinants of health ICD-10 codes (Z-codes). Boldface indicates statistical significance (P < .05).

Abbreviations: aOR, adjusted odds ratio; ICD, International Classification of Diseases; ref, reference.

The Data Supplement (Fig S1) displays the distribution of regional hospital visits across the 10 most frequently visited hospital systems by both county safety-net and academic medical center patients. Among county safety-net patients, 5.4% (2,278/42,390) of hospital encounters at Parkland had an associated Z-code. County safety-net primary patients were more likely to have SDOH identified during a county safety-net hospital visit compared with academic medical center primary patients during an academic medical center visit (P < .001). Patients who had an associated Z-code entry were more likely to seek care at more than one hospital system (v those without Z-code, 0.4% v 0.1%, P = .01).

DISCUSSION

In our population-based sample of adults with incident cancer and their comprehensive hospital visits, we noted a rise in Z-code use from 1% to over 11% of hospital visits by 2023. This increase in Z-code use is likely related to the uptake of SDOH as it began to be required by 2024.⁸ Providers may have begun to more frequently incorporate Z-codes as billing codes to document more complex care and decision making for increased reimbursement.¹⁹ Across all hospital systems in the region, we also demonstrated that patients with hospital visits at the county system were more likely to have an associated Z-code entry compared with other hospital systems.

Providers at the safety-net system are aware that they serve a largely uninsured patient population and may be more cognizant of the importance in documenting SDOH. Many of the uninsured patients are only able to obtain payment for medical care at the safety-net through its robust Indigent care coverage program.²⁰ In our study, we highlight that over 60% of hospital visits made by county safety-net patients were to Parkland Hospital. In comparison, the academic medical center patient population visited a broader array of regional hospitals and hospital systems. This may explain why a higher proportion of county hospital encounters had an associated Z-code when compared with academic medical center encounters.

Our analysis demonstrated that patients within the county health care system were significantly more likely to have a Z-code entered, even after adjusting for sociodemographic characteristics. This finding suggests that institutional practices and workflow changes may have contributed to the observed increase in Z-code utilization. Beginning in late 2022, the county safety-net system implemented several initiatives to enhance SDOH identification and documentation. These included routine integration of SDOH screening into outpatient clinic intake workflows, often conducted by medical assistants during patient visits. During this time, Best Practice Alerts were also introduced within the EMR to notify providers of potential unmet SDOH needs. These alerts prompted consideration of appropriate Z-code assignment and encouraged referral to social work services when indicated. Collectively, these system-level interventions likely improved both the identification of patient social needs and increased Z-code documentation.

The safety-net hospital system did not participate in the Accountable Care Organization Realizing Equity, Access, and Community Health (ACO REACH) pilot program by the CMS, which encouraged participants to collect and report SDOH data. The academic practice is part of a larger ACO and did participate in ACO REACH. However, since the SDOH screening remained only as an encouraged factor, but not required until 2024, the academic practice had not yet instituted a larger systematic encounter-level screening process to capture beneficiary-level SDOH needs screening. As SDOH screening becomes more prevalent and required, future research analyzing the impact of these programs would be of great interest.

Currently, Z-codes are not incorporated into the CMS Hierarchical Condition Category (HCC) risk-adjustment model used for Medicare Advantage payment.²¹ Z-code entries and SDOH diagnoses do not currently raise a patient’s calculated risk score. Social complexity may eventually be incorporated into risk-adjustment or value-based payment methodologies, like how HCCs adjust for medical comorbidities. If trends in documentation practices continue to improve and evolve, Z-code data may potentially offer a feasible foundation for incorporating social complexity into risk-adjustment or value-based payment models. In this case, further work centered on validating the accuracy of Z-code entry at the patient level may be needed before it could support payment decisions.

We found that housing problems was the most frequently documented needs within our cohort, seen in approximately 3% of our patients. Prior studies analyzing hospital use of Z-code entries for Medicare fee-for-service beneficiaries also document housing as the most frequently used Z-code.⁷ Housing insecurity and homelessness is a social need faced by an estimated 9% of patients with newly diagnosed cancer and is associated with poorer outcomes within cancer care.^22,23 As Z-codes that describe homelessness have recently been designated from noncomorbid conditions to comorbid condition in October 2023, there may be further increase in the use of Z59, which represents housing and economic circumstances, to describe homelessness for reimbursement.²⁴ This change may help better balance increased costs associated with unhoused patients who may require more resources during hospitalizations.²⁵

Our study identified various patient populations who may be at greater risk of having unmet health-related social needs across large urban geographic region. Patients who identified as Black as well as uninsured patients were more likely to have a Z-code entered. This has similarly been shown in studies suggesting that non-White and uninsured patients are more likely to have Z-codes assigned.²⁶ Patients without insurance have been shown to be disproportionately affected by SDOH and often have lower socioeconomic status and educational status.^27,28 Likewise, Hispanic patients have reported higher unmet needs, particularly food insecurity and financial strain, which have negatively affected their chronic disease burden.^29,30 Although it is possible these patient populations have increased unmet SDOH needs, these patients may also have Z-codes assigned due to underlying bias in documentation.²¹ For instance, Black patients may often be described using stigmatizing language or descriptors in the EMR, which may also affect how Z-codes are documented.^31,32

In our study, 23% of patients with Z-codes were diagnosed with gastrointestinal or colon cancer, and a higher proportion of patients with Z-codes were diagnosed with advanced stage cancer compared with those without any Z-code. These findings may reflect prior findings that transportation needs, housing insecurity, and food insecurity are often found in patients diagnosed with gastrointestinal cancers. In particular, financial toxicity and strain has been shown to be highly prevalent with treatment of gastrointestinal cancers and is associated with poorer outcomes and quality of life.^33,34 Similarly, patients who present with late-stage, advanced cancer diagnoses often have unmet SDOH needs, which contribute to delayed diagnosis and treatment.^10,35 As a result, it is increasingly important to screen patients diagnosed with cancer for unmet SDOH, which is a driver for poor survival across many types of cancers.³⁶

Our findings have several important implications for oncology care delivery. First, information derived from patterns of Z-code entry may help identify gaps in SDOH screening and improve screening workflows. Examining Z-code data within a program can also help with resource allocation, including expansion of social work, patient navigation, and community-based support services to address commonly identified needs for a patient population. Finally, integration of Z-code documentation into routine oncology practice may facilitate more targeted interventions, such as referrals to supportive care services, transportation assistance, or financial counseling, which could help improve access to treatment and continuity of care. In effect, this will allow for personalization of the development and deployment of SDOH interventions for patients to further improve the effectiveness of oncology care delivery.

Although this study is, to our knowledge, the first to examine SDOH Z-code entry among a diverse, population-based sample of adults with cancer and their comprehensive hospital visits, our findings should be interpreted with some limitations. We did not have information regarding Z-code entry beyond 2023, so we are unable to further characterize trends after the integration of Z-codes after 2024. Additionally, Z-code usage may not fully capture patients’ underlying social needs, as its use depends on the presence, quality, and consistency of screening and documentation practices, which varies across institutions. We also only capture Z-codes among those patients who visited the hospital and were screened for this, and we cannot describe how accurately the screening and coding has captured underlying SDOH problems faced by patients. Likewise, our data set does not fully account for SDOH screening or documentation occurring in outpatient oncology or primary care settings if a patient did not subsequently have an emergency department or hospital visit.

In conclusion, we illustrate how Z-code entry is variable across different hospital systems despite a gradual rise in Z-code usage from 2015 to 2023. At the same time, overall Z-code use remains low across hospital systems, underscoring the gap between policy goals and real-world practice. As new policies and incentives are developed, they will help assess whether these system-level changes can further increase Z-code uptake. Further work is needed to understand patient and provider factors influencing Z-code entry, including physician attitudes toward SDOH screening and whether they propagate biases into the EMR. As delivery of cancer care is multifaceted and complex, proper documentation of SDOH needs through Z-codes can be a powerful tool for physicians to identify and monitor patient needs.

Supplementary Material

Data Supplement

CONTEXT.

Key Objective

What is the extent and breadth of social determinants of health diagnosis code usage (Z-code) among patients with cancer?

Knowledge Generated

In a population-based sample of adults with cancer and their comprehensive hospital visits, Z-code use increased to over 11% of hospital visits by 2023, and housing and economic problems accounted for half of the Z-codes. Patients treated at safety-net hospitals had nearly three times higher Z-code use than those at academic medical centers.

Relevance

Improving identification and documentation of these nonmedical drivers of health is foundational to developing and evaluating interventions to alleviate them and improve clinical care delivery.