Social determinants of health diagnosis codes in in-person and telehealth visits during pregnancy in the United States, 2020–2025: a repeated cross-sectional and case-control study

Abstract

Background

Social determinants of health (SDoH) diagnosis codes (also known as Z-codes) are used to document patients’ SDoH, which could affect a patient’s health. However, their use during pregnancy is unknown. Additionally, with the growth in telehealth use in recent years, how the Z-codes are documented across in-person outpatient/office and telehealth prenatal visits is undetermined. Therefore, the objective of this study was to document trends in prenatal Z-code diagnosis in telehealth (audio-only and audio-visual) and in-person outpatient/office visits, and the association between visit type and Z-codes during pregnancy.

Methods

Repeated cross-sectional and case-control designs were employed using the Epic Cosmos database (July 2019–March 2025). All pregnancies with start dates on or after July 01 2020 and visits until March 31, 2025 were considered. Total in-person outpatient/office and telehealth visits (denominators) were summed, and visits with a Z-code were determined (numerators). Rates were expressed per 1,000 visits. Case-control analyses, stratified by provider type, were conducted to estimate the relationship between visit type and Z-code incidence. The first documentation of Z-codes (a wash-out period of 365 days) in each pregnancy episode was identified (case visits), and control visits were gathered. Multivariable logistic regressions were estimated and odds ratios (OR) were reported.

Results

A total of 3,435,577 pregnancies were included, totaling 33,376,363 in-person outpatient and telehealth visits. Overall average Z-code rates per 1,000 visits were: 8.64 (in-person), 6.71 (audio-only), and 12.35 (audio-visual). The rates increased on average for all three visit types over the study period, with rates slightly higher for audio-visual visits. In the provider sub-group of family practice/ob-gyn/other clinicians, audio-visual visits had no significant association with Z-code incident diagnosis (OR: 1.04 [0.99; 1.10]), while audio-only visits had lower odds (OR: 0.60 [0.47; 0.78]), compared to in-person visits. Among nursing/midwife provider sub-group, audio-visual visits were associated with 42% higher odds (OR: 1.42 [1.35; 1.49]), while audio-only visits were associated with 48% lower odds (OR: 0.52 [0.43; 0.63]), compared to in-person visits.

Conclusion

Audio-visual visits with nurses/midwives were better at capturing SDoH measures. The integration of audio-visual visits with these providers for routine follow-up during pregnancy could better capture patients’ SDoH.

Background

The International Classification of Diseases-10-Clinical Modification (ICD-10-CM) diagnosis codes for social determinants of health (SDoH) were released in 2015 to capture factors outside the healthcare system that could affect a patient’s health. [1] Given that these diagnosis codes, also known as Z-codes, reflect patients’ SDoH, they could help understand disease complications, treatment non-adherence or discontinuation, and lack of or excess healthcare utilization. [2–4] For example, primary care patients with at least one social need had lower odds of cancer screening, chronic disease management, and use of any healthcare service. [4] A meta-analysis reported that patients with food insecurity and housing instability had 44% and 36% lower odds of medication adherence, respectively. [5] When an overall SDoH was considered, the odds of medication adherence were 25% lower among such patients. [5]

These Z codes have been used by providers and payers (commercial and Medicare/Medicaid) for billing purposes since 2015. [6] A study using United States (US) national surgical data from 2015 to 2021 found that only 3.2% of surgeries had an SDoH Z-code, [7] suggesting that SDoH codes are likely underutilized. Another study reported that 2.0% of patients and 2.7 per 1,000 visits had SDoH Z-codes. [8] A mixed-methods study found that the rate of Z-code documentation in the medical records between 2015 and 2020 (~ 1% of patients) did not correspond with the burden of SDoH in the patients. [9] Interviews with the providers revealed barriers to documenting these codes, which included unease around broaching social needs of the patients and limited time availability to discuss them, which likely explains their under-coding. [9] Although the utilization rates of Z-codes have been documented among patients with different clinical (hypertension, hospitalized patients) and socio-demographic (different insurance types such as Medicare, Medicaid in the US) profiles, [10–13] it is unknown how these Z-codes are used during pregnancy. Prior studies have described the relationships between pregnant women’s SDoH and adverse obstetric outcomes. [14, 15] Also, understanding how these Z-codes are documented across audio-only and audio-visual telehealth modalities can help assess the current landscape of their use and inform efforts to increase their use to better match actual social needs.

This study had two objectives. First, we investigated the rates of Z-codes documented in in-person and telehealth visits overall and stratified by rural/urban residence, race/ethnicity, gestational weeks, social vulnerability index (SVI) quartiles, and provider type (specialty). Second, we quantified the association between the visit modality and the incident Z-code documentation separately for provider type while also adjusting for covariates that could potentially confound the relationship. The findings from this study could help determine the visit modality that can address patients’ social needs and ultimately contribute to the design of hybrid prenatal care models.

Methods

Study design and data source

We used a repeated cross-sectional study design for the trends and a case-control study design for the relationship between the visit type and the incident Z-code diagnosis. Data used in this study came from Epic Cosmos, a dataset created in collaboration with a community of Epic health systems representing more than 290 million patient records from over 1,633 hospitals and 37,900 clinics from all 50 US states, D.C., Canada, Lebanon, and Saudi Arabia. [16] The database is longitudinal and has information on patient demographics, healthcare encounters, diagnoses, procedures, medications administered and dispensed, and laboratory results. We used the data from July 01, 2019 and March 31, 2025, and restricted the study sample to individuals residing in the US.

Study population and sample selection

For the first objective (repeated cross-sectional analysis), we first identified all active pregnancies in the database with estimated start dates on or after July 1, 2020, using the PregnancyFact table of the database. A one-year pre-pregnancy period was required for all women in the sample: at least one healthcare encounter was required in that period to ensure that the women were active in the database. We gathered all in-person and telehealth visits from the prenatal period until March 31, 2025, which was the latest available data at the time of the analysis (administrative censoring). Pregnancies were restricted to patients aged 15–50 years at the start of pregnancy. Pregnancies with gestational age > 42 weeks, which could be data error, were excluded. Multiple pregnancies for the same individual were considered, and all delivery outcomes during the study period were included. Pregnancies with duplicate start and/or end dates were excluded. In-person outpatient/office visits and telehealth visits between the start and end dates of the pregnancies were identified.

For the second objective (case-control analysis), we estimated the relationship between the visit modality and the first documentation of the Z-codes. As diagnoses are carried forward in electronic medical records, cross-sectional rates of the diagnosis across visit types may not reflect the true relationship between visit type and Z-codes, thereby necessitating a longitudinal study design. To that end, we first identified all visits with the first (incident) documentation of Z-codes in the respective pregnancy episode. To establish that the pregnancy episodes did not have a history of Z-codes, we excluded the episodes that had a Z-code in the 365 days before the incident diagnosis. The visit with the incident Z-code diagnosis for each pregnancy episode was considered the case visit. For the controls, which were visits that did not include a Z code, we gathered all in-person outpatient and telehealth visits during pregnancy (from those before the case visits for those with a case visit; all visits for those without a case visit). This approach was adopted because those prior visits could potentially have captured pregnant patients’ SDoH, thereby meeting one of the requirements for a case-control study. We then stratified all visits (both case and control visits) based on the type of provider associated with the visits: (1) Family practice/obstetrician-gynecologist (ob-gyn)/other clinicians, (2) Nursing/midwife professionals, (3) Other clinical professionals, and (4) Missing provider type. We combined multiple provider specialties to create three categories and a missing category, as more granular categories would lead to a smaller sample size, particularly for telehealth visits. Because an ob-gyn visit, in which laboratory tests are performed, may not prioritize SDoH, while a social worker visit may be more dedicated to SDoH assessment, we compared in-person outpatient/office and telehealth visits in the same provider category.

Study measures

In-person outpatient/office visits and telehealth visits between the start and end dates of the pregnancies were identified. In-person outpatient/office visits were identified using the encounter type. Telehealth visits were determined using encounter and Current Procedural Terminology (CPT) codes: Audio-visual: Encounter type = “Telemedicine” and CPT code is not NULL; Audio-only: (Encounter type in (“Telephone Visit” or “Telephone”) and CPT code is not NULL), OR (other Encounter type and CPT code in (“99441”, “99442”, “99443”)). Each visit (both in-person and telehealth) was examined for the presence of SDoH ICD-10-CM diagnosis codes. Following Z-codes were considered (first three digits): Z55, Z56, Z57, Z59, Z60, Z62, Z63, Z64, and Z65. These Z-codes were identified from the DiagnosisFact table of the database. All types of diagnoses (billing, encounter, secondary) were considered. Telehealth visits were further divided into audio-only and audio-visual visits using the encounter types and CPT codes. Visits were excluded if they were flagged as both telehealth and in-person visits (likely an error). All visits were divided into calendar months, and monthly trends per 1,000 visits were calculated for in-person, audio-only, and audio-visual visits. We also conducted stratified analysis for in-person versus telehealth visits (combined audio-only and audio-visual) based on race/ethnicity, rural/urban residence, estimated gestational weeks at the respective visit, provider type, and census tract-level SVI [17] based on the patients’ residence. Race/Ethnicity was identified from the PatientDim table and divided into: (1) Non-Hispanic White, (2) Non-Hispanic Black, (3) Hispanic, and (4) Other/Missing. We excluded the Other/Missing category for the race-stratified analysis. Rural/Urban residence was determined using census tract-level Rural-Urban Commuting Area (RUCA) codes, with values 1–3 denoting an urban area and 4–10 denoting a rural area. Patients with missing RUCA information were excluded from the rural/urban-specific analysis. Estimated gestational weeks were divided into three groups, reflecting the three trimesters: weeks 1–13, weeks 14–27, and weeks 28–end of pregnancy. SVI is a census-tract level score created by the US Centers for Disease Control and Prevention (CDC) to rank census tracts based on socioeconomic status, household characteristics, transportation, and racial/ethnic minority status, which can help identify regions that are prone to worse outcomes in the event of disasters or disease pandemics. [17] The overall SVI score was divided into four quartiles (1st quartile: least vulnerable; 4th quartile: most vulnerable).

For the case-control analysis, we also gathered covariates for all case and control visits from the previous 365 days. The following covariates were considered: asthma, chronic kidney disease, ischemic heart disease, alcohol use disorder, tobacco use disorder, other substance use disorder, pre-existing diabetes, pre-existing hypertension, obesity, gestational diabetes, gestational hypertension, mild pre-eclampsia, severe pre-eclampsia, previous cesarean delivery, and multiple gestation (Appendix Table). Age at the start of pregnancy, estimated gestational weeks at each visit, rural/urban residence, and race/ethnicity were considered as socio-demographic covariates.

Statistical analyses

The monthly rates of Z-code documentation were calculated per 1,000 visits, separately by the visit modality. For the case-control analysis, we estimated multivariable logistic regressions using a binary telehealth use variable as the exposure of interest and the previously listed covariates as other independent variables in the regression, separately for each provider type. Additionally, we fitted multivariable logistic regressions, stratified by provider type, using an exposure variable with three groups: in-person, audio-only, and audio-visual. Odds ratios (OR) and 95% confidence intervals (CI) are reported. Standard errors were clustered at the patient level. Microsoft SQL Server and R/RStudio were used for data extraction and statistical analysis, respectively.

Sensitivity analyses

We conducted two sensitivity analyses. First, we restricted the visit-level sample to visits within 20 weeks of gestation. As visits in the latter stages of a pregnancy are focused on maternal-fetal care, including physical and physiological monitoring, they may be less relevant for capturing SDoH. Also, high-risk pregnancies, which would warrant more maternal-fetal monitoring and likely less emphasis on SDoH diagnosis, are determined by conditions that are diagnosed after 20 weeks of gestation, such as gestational diabetes, pre-eclampsia/eclampsia. For this analysis, similar to the main analysis, we performed multivariable logistic regressions, stratifying by provider type.

Second, we conducted a case-crossover study, [18] in which we analyzed the sample of pregnancy episodes with the SDoH Z code (cases). We then extracted all the visits for these pregnancy episodes that occurred before the case visit, similar to the main analysis (but here, restricted to those who ultimately had the Z code), which served as control visits. The assumption for this analysis is that for pregnant women who had an SDoH Z code, all prior visits in that pregnancy episode could have documented the Z code.

Study results

A total of 12 million pregnancies were observed in the database. After applying all exclusion criteria, 3,435,577 pregnancies were included in the final sample, which had 33,376,363 in-person outpatient or telehealth visits in the study period (Fig. 1). When the Z-code rates per 1,000 visits were estimated by in-person outpatient/office and telehealth visits, the overall rates were 8.64 and 11.90, respectively (Table 1). When the telehealth visits were divided into audio-only and audio-visual visits, the rates were, respectively, 6.71 and 12.35 per 1,000 visits.

Fig. 1.

Sample selection diagram for pregnancies for the trend analyses

Table 1.

Overall rates of Social Determinants of Health (SDoH) Z codes per 1,000 visits during the prenatal period (July 2020–March 2025) among pregnant women aged 15–50 years

Groups	In-person outpatient/office visits	Telehealth visits
Overall rates	8.64	11.90 (any) 6.71 (audio-only) 12.35 (audio-visual)
Race/Ethnicity
Non-Hispanic White	5.03	9.04
Non-Hispanic Black	13.70	18.82
Hispanic	17.23	15.75
Rural/Urban residence
Rural	5.55	7.55
Urban	9.20	12.38
Gestational weeks at the visit
Weeks 1–13	8.62	13.43
Weeks 14–27	8.79	11.69
Weeks 28–pregnancy end	8.55	10.23
SVI quartiles
1st quartile (least vulnerable)	4.41	7.55
2nd quartile	6.49	9.43
3rd quartile	9.44	12.91
4th quartile (most vulnerable)	14.31	17.12
Provider type
Family practice/Ob-gyn/Other clinicians	7.61	10.03
Nursing/Midwife professionals	12.38	12.36
Other clinical professionals	13.07	14.77
Missing provider type	3.75	6.99

Figure 2 shows the trends in Z-code rates for in-person, audio-only, and audio-visual visits. The Z-code rates were consistently higher for audio-visual visits compared to in-person and audio-only visits. The Z-code rates increased for all three visit types during the study period. The rates in March 2025 (the last month of the study period) were: 10.99 (in-person), 10.30 (audio-only), and 13.27 (audio-visual).

Fig. 2.

Rates of Social Determinants of Health (SDoH) Z codes per 1,000 visits during the prenatal period (July 2020–March 2025) among pregnant women aged 15–50 years for In-person, Audio-only, and Audio-visual visits. Notes: In-person visits were limited to outpatient/office visits; audio-visual and audio-only visits were identified using a combination of encounter types and procedure codes

Figure 3 shows the Z-code trends for in-person and telehealth visits by race/ethnicity. The Z-code rates were higher for Hispanic and non-Hispanic Black women compared to non-Hispanic White patients. Among non-Hispanic White and Black patients, telehealth visits had higher rates of Z-code documentation compared to in-person visits, while in-person visits had higher rates among Hispanic patients. The average rates across the racial-ethnic groups over the study period were: non-Hispanic White: 5.03 (in-person), 9.04 (telehealth); non-Hispanic Black: 13.70 (in-person), 18.82 (telehealth), and Hispanic: 17.23 (in-person), 15.75 (telehealth) (Table 1).

Fig. 3.

Rates of Social Determinants of Health (SDoH) Z codes per 1,000 visits for telehealth and in-person visits during the prenatal period (July 2020–March 2025) among pregnant women aged 15–50 years, stratified by race/ethnicity. Notes: In-person visits were limited to outpatient/office visits; telehealth visits were identified using a combination of encounter types and procedure codes. Race/Ethnicity was determined using PatientDim table in the database

Figure 4 reports the trends by rural/urban residence. The Z code rates were higher among urban residents than among rural residents, and in telehealth visits (compared to in-person) for both rural and urban regions during the study period. The average rates across rural/urban groups over the study period were: Rural: 5.55 (in-person), 7.55 (telehealth); Urban: 9.20 (in-person), 12.38 (telehealth) (Table 1).

Fig. 4.

Rates of Social Determinants of Health (SDoH) Z codes per 1,000 visits for telehealth and in-person visits during the prenatal period (July 2020–March 2025), stratified by rural/urban region of residence. Notes: In-person visits were limited to outpatient/office visits; telehealth visits were identified using a combination of encounter types and procedure codes. Rural/Urban residence was identified using Rural-Urban Commuting Area Codes: 1–3 were categorized as Urban, and 4–10 were Rural

The rates were higher for telehealth visits across all trimesters, although the differences were much larger for first and second trimesters (Fig. 5). The overall average rates were: weeks 1–13: 8.62 (in-person), 13.43 (telehealth); weeks 14–27: 8.79 (in-person), 11.69 (telehealth), and weeks 28–pregnancy end: 8.55 (in-person), 10.23 (telehealth) (Table 1).

Fig. 5.

Rates of Social Determinants of Health (SDoH) Z codes per 1,000 visits for telehealth and in-person visits during the prenatal period (July 2020–March 2025), stratified by estimated gestational weeks of pregnancy. Notes: In-person visits were limited to outpatient/office visits; telehealth visits were identified using a combination of encounter types and procedure codes. The trend lines were truncated by the end of November 2024 for gestational weeks 1–13 and February 2025 for gestational weeks 14–27 because the latest pregnancy start date in the data was August 31, 2024

The overall average rates across SVI quartiles were: 1st quartile: 4.41 (in-person), 7.55 (telehealth); 2nd quartile: 6.49 (in-person), 9.43 (telehealth); 3rd quartile: 9.44 (in-person), 12.91 (telehealth); and 4th quartile: 14.31 (in-person), 17.12 (telehealth) (Table 1). Figure 6 shows the Z-code rates for SVI quartiles: the rates were higher for telehealth visits than for in-person visits in all SVI quartiles.

Fig. 6.

Rates of Social Determinants of Health (SDoH) Z codes per 1,000 visits for telehealth and in-person visits during the prenatal period (July 2020–March 2025), stratified by census tract-level social vulnerability index (SVI). Notes: In-person visits were limited to outpatient/office visits; telehealth visits were identified using a combination of encounter types and procedure codes. The overall SVI score was divided into four quartiles, with 1st quartile indicating the least vulnerable census tracts and 4th quartile indicating the most vulnerable

The overall average Z-code rates across provider categories were: family practice/ob-gyn/other clinicians: 7.61 (in-person), 10.03 (telehealth); nursing/midwife professionals: 12.38 (in-person), 12.36 (telehealth); other clinical professionals: 13.07 (in-person), 14.77 (telehealth); and missing provider type: 3.75 (in-person), 6.99 (telehealth) (Table 1). Figure 7 shows the trends by provider type: there was more variability between in-person and telehealth visits across the provider type.

Fig. 7.

Rates of Social Determinants of Health (SDoH) Z codes per 1,000 visits for telehealth and in-person visits during the prenatal period (July 2020–March 2025), stratified by provider type. Notes: In-person visits were limited to outpatient/office visits; telehealth visits were identified using a combination of encounter types and procedure codes

For the case-control analysis, 105,982 case visits (visits with the incident Z code diagnosis ) and 32,534,708 control visits (visits without the incident Z code) were determined (Table 2). Among the cases, 4.75% were telehealth visits, while 3.00% of the controls were telehealth visits. In the stratified multivariable analyses, telehealth visits were not associated with the diagnosis among the family practice/ob-gyn/other clinicians (OR: 1.02 [0.97; 1.07]), whereas telehealth visits had a positive association among nursing/midwife professionals (OR: 1.29 [1.22; 1.35]) (Table 3). The relationships were drastically different when telehealth was divided into audio-only and audio-visual visits. Among nursing/midwife professionals, audio-only visits had lower odds of the incident Z-code (OR: 0.52 [0.43; 0.63]), while audio-visual visits had much higher odds (OR: 1.42 [1.35; 1.49]), compared to in-person outpatient/office visits. Among family practice/ob-gyn/other clinicians, audio-only visits had much lower odds (OR: 0.60 [0.47; 0.78]) than in-person visits, while audio-visual visits had similar odds (OR: 1.04 [0.99; 1.10]) relative to in-person visits. The results were consistent in the sensitivity analysis for nursing/midwife professionals (Tables 3 and 4); however, audio-visual visits had slightly lower odds of Z-code diagnosis compared to in-person visits in the sub-group of family practice/ob-gyn/other clinicians in the case-crossover analysis (OR: 0.87 [0.83; 0.92]).

Table 2.

Characteristics of prenatal visits based on the outcome (incident diagnosis of Z-codes)

Characteristics	Outcome = 0; N = 32,534,708; n (%)	Outcome = 1; N = 105,982; n (%)
Visit type
In-person visit	31,559,211 (97.00)	100,950 (95.25)
Telehealth visit	975,497 (3.00)	5,032 (4.75)
Age at start of pregnancy (years): mean (sd)	29.83 (5.72)	29.06 (6.21)
Race/Ethnicity
Non-Hispanic White	16,422,262 (50.48)	33,371 (31.49)
Non-Hispanic Black	4,841,382 (14.88)	28,309 (26.71)
Hispanic	4,302,434 (13.22)	24,914 (23.51)
Other Race	2,138,402 (6.57)	7,713 (7.28)
Missing	4,830,228 (14.85)	11,675 (11.02)
Rural/Urban region of residence
Rural/Missing	5,117,266 (15.73)	11,064 (10.44)
Urban	27,417,442 (84.27)	94,918 (89.56)
Provider type
Family practice/ob-gyn/other clinicians	17,643,967 (54.23)	51,633 (48.72)
Nursing/midwife professionals	5,409,212 (16.63)	26,918 (25.40)
Other clinical professionals	4,894,896 (15.05)	21,496 (20.28)
Missing provider type	4,586,633 (14.10)	5,935 (5.60)
Asthma	1,975,209 (6.07)	6,696 (6.32)
Chronic renal disease	211,680 (0.65)	493 (0.47)
Ischemic heart disease	48,250 (0.15)	140 (0.13)
Pre-existing diabetes	1,018,495 (3.13)	2,810 (2.65)
Pre-existing hypertension	2,412,809 (7.42)	5,709 (5.39)
Obesity	14,233,096 (43.75)	37,982 (35.84)
Alcohol use disorder	97,786 (0.30)	622 (0.59)
Tobacco use disorder	1,166,696 (3.59)	6,283 (5.93)
Other substance use disorder	1,637,075 (5.03)	8,817 (8.32)
Gestational diabetes	2,913,508 (8.96)	5,202 (4.91)
Gestational hypertension	634,247 (1.95)	1,029 (0.97)
Multiple gestation	673,700 (2.07)	1,244 (1.17)
Previous Cesarean delivery	2,786,954 (8.57)	6,805 (6.42)
Mild Pre-eclampsia	212,719 (0.65)	364 (0.34)
Severe Pre-eclampsia	52,875 (0.16)	132 (0.12)
Number of gestational weeks at each visit: mean (sd)	24.09 (11.09)	18.42 (10.84)

Sd Standard deviation

Table 3.

Multivariable logistic regression results for the outcome of Z diagnosis code for social determinants of health (SDoH)

Exposure (visit type) Ref = in-person outpatient visit	Provider type (sub-groups) Odds ratio (95% confidence interval)
	Family practice/Ob-gyn/Other clinicians	Nursing/Midwife professionals	Other clinical professionals	Missing provider type
Any telehealth	1.02 (0.97; 1.07)	1.29 (1.22; 1.35)	0.96 (0.91; 1.01)	2.88 (2.51; 3.31)
Audio-only	0.60 (0.47; 0.78)	0.52 (0.43; 0.63)	0.82 (0.66; 1.01)	2.32 (1.52; 3.54)
Audio-visual	1.04 (0.99; 1.10)	1.42 (1.35; 1.49)	0.97 (0.92; 1.03)	2.96 (2.56; 3.43)
Sensitivity analyses
Visits restricted to first 20 weeks of gestation
Any telehealth	0.91 (0.86; 0.97)	1.41 (1.33; 1.49)	0.92 (0.87; 0.99)	3.45 (2.97; 4.02)
Audio-only	0.46 (0.33; 0.63)	0.58 (0.45; 0.75)	0.73 (0.57; 0.95)	2.98 (1.93; 4.61)
Audio-visual	0.94 (0.89; 1.00)	1.51 (1.43; 1.60)	0.94 (0.88; 1.01)	3.52 (3.00; 4.13)
Visits restricted to those episodes with the SDoH Z code (case-crossover)
Any telehealth	0.86 (0.81; 0.90)	1.15 (1.09; 1.21)	0.84 (0.80; 0.90)	2.51 (2.06; 2.91)
Audio-only	0.60 (0.46; 0.77)	0.54 (0.44; 0.66)	0.67 (0.53; 0.84)	3.84 (2.46; 6.01)
Audio-visual	0.87 (0.83; 0.92)	1.23 (1.17; 1.30)	0.86 (0.81; 0.91)	2.42 (2.07; 2.83)

Table 4.

Characteristics of prenatal visits based on the outcome (incident diagnosis of Z-codes), limiting the visits to first 20 weeks of pregnancy

Characteristics	Outcome = 0; N = 11,946,587; n (%)	Outcome = 1; N = 64,296; n (%)
Visit type
In-person visit	11,420,893 (95.60)	60,604 (94.26)
Telehealth visit	525,694 (4.40)	3,692 (5.74)
Age at start of pregnancy (years): mean (sd)	29.93 (5.70)	29.09 (6.21)
Race/Ethnicity
Non-Hispanic White	6,072,480 (50.83)	20,363 (31.67)
Non-Hispanic Black	1,757,082 (14.71)	17,261 (26.85)
Hispanic	1,589,281 (13.30)	15,094 (23.48)
Other Race	781,257 (6.54)	4,643 (7.22)
Missing	1,746,487 (14.62)	6,935 (10.79)
Rural/Urban region of residence
Rural/Missing	1,859,663 (15.57)	6,509 (10.12)
Urban	10,086,924 (84.43)	57,787 (89.88)
Provider type
Family practice/ob-gyn/other clinicians	5,772,637 (48.32)	30,554 (47.52)
Nursing/midwife professionals	2,174,137 (18.20)	16,791 (26.12)
Other clinical professionals	1,987,325 (16.64)	13,212 (20.55)
Missing provider type	2,012,488 (16.85)	3,739 (5.82)
Asthma	607,688 (5.09)	3,100 (4.82)
Chronic renal disease	47,536 (0.40)	184 (0.29)
Ischemic heart disease	14,100 (0.12)	65 (0.10)
Pre-existing diabetes	352,078 (2.95)	1,480 (2.30)
Pre-existing hypertension	597,546 (5.00)	2,440 (3.79)
Obesity	3,307,084 (27.68)	15,814 (24.60)
Alcohol use disorder	43,196 (0.36)	357 (0.56)
Tobacco use disorder	422,516 (3.54)	3,227 (5.02)
Other substance use disorder	567,294 (4.75)	4,363 (6.79)
Gestational diabetes	336,161 (2.81)	1,280 (1.99)
Gestational hypertension	38,548 (0.32)	164 (0.26)
Multiple gestation	184,587 (1.55)	538 (0.84)
Previous Cesarean delivery	490,486 (4.11)	1,934 (3.01)
Mild Pre-eclampsia	18,098 (0.15)	67 (0.10)
Severe Pre-eclampsia	6,650 (0.06)	37 (0.06)
Number of gestational weeks at each visit: mean (sd)	11.39 (5.09)	10.69 (4.46)

sd Standard deviation

Discussion

The overall documentation rates of Z diagnosis codes were low during pregnancy, indicating their under-utilization. We found that Z-code rates were higher for telehealth visits than for in-person visits on average among pregnant women during the prenatal period, driven by higher rates in audio-visual visits. The Z-code rates were lower for audio-only visits compared to in-person visits, although considerable variation was observed in the rates for audio-only visits over the study duration. The higher rates of Z-codes in telehealth visits were observed among both urban and rural residents, two out of the three racial/ethnic groups, all trimesters, residents in all four quartiles of SVI, and the provider type of family practice/ob-gyn/other clinicians. However, in the case-control analysis, which was focused on estimating the relationship between the visit type and incident Z-code diagnosis, found that telehealth visits, particularly audio-visual visits, had higher odds of documenting the Z-code relative to in-person outpatient/office visits among nursing/midwife professionals but not among the family practice/ob-gyn/other clinicians category. The sensitivity analyses were consistent with the main analyses, suggesting that telehealth nursing/midwife visits were more likely to capture SDoH diagnosis than in-person nursing/midwife visits even during the early stages of pregnancy.

Telehealth visits have been labeled as “virtual house calls”, with the advantage of navigating a patient’s residence. [19] A qualitative study found that physicians appreciated telehealth visits as they could see the patient’s environment, including support systems (family members, pets) and potential hazardous exposures. [20] The clinicians also revealed that the visits aided with services such as medication reconciliation. [20] Another qualitative study reported similar findings: through telehealth visits, physicians interacted with family members and found additional information about the patient’s lifestyle and health behaviors. [21] A study of prenatal visits in nurse-midwifery clinics in 2019–2021 found that total prenatal visits increased in 2021 compared to 2019, driven by an increase in telehealth visits. [22] Patients have also reported higher satisfaction with nurse-midwife telehealth visits. [23] Our findings of higher Z-code documentation in audio-visual visits with nursing/midwife professionals are congruent with these prior studies, which suggest that telehealth visits may reveal the SDoH issues. Although concerns exist with telehealth visits during pregnancy, such as difficulty monitoring fetal growth and/or the mother’s vital signs, and lack of private space for patients,²³ telehealth visits for routine follow-up could better capture patients’ social needs.

A comprehensive recording of SDoH and its subsequent use in predictive models has led to better prediction of several clinical outcomes, such as mortality, all-cause hospitalization, cardiovascular hospitalization, emergency department visits, and healthcare costs. [24, 25] Improving the prediction of these utilization and clinical outcomes can aid in tailoring prevention and treatment strategies to patients’ needs. Incorporating SDoH in risk-adjustment models for plan payment could also improve insurance payments for underserved patients. [26] Given that audio-visual visits are more likely to capture patients’ SDoH, a hybrid prenatal care model can likely result in better recording of SDoH and ultimately improvement in clinical outcomes.

Our study has at least four limitations. First, Epic Cosmos only contains information from organizations that use Epic EMR and contribute data to the Cosmos database; in-person and/or telehealth visits at non-Epic organizations would be missing in our study database and could affect the Z-code rates if the missingness is not random. Second, the diagnosis rates of Z-code do not necessarily correspond to SDoH rates in the patient population and may be differentially represented owing to patient-, clinician-, or health system- factors. Third, missing data for RUCA, race/ethnicity, SVI, and provider type could affect an accurate estimation of the rates. Fourth, unmeasured confounding could be present and could have affected the adjusted relationship between visit type and Z-code diagnosis: patient-level factors such as availability of telehealth services, internet access, and digital literacy of the patients were not accounted for. Additionally, health system factors such as the adoption (or lack thereof) of SDoH screening, and providers responsible for such screening could systematically bias the estimated relationship between the visit type and Z-code diagnosis.

In conclusion, we found that audio-visual telehealth modality during prenatal period was more likely to capture SDoH of pregnant patients than in-person visits among visits with nursing/midwife professionals, while no such relationship or slightly lower odds was observed for visits with family practice/ob-gyn/other clinicians. Audio-only visits were least likely to capture the Z codes, underscoring that a binary telehealth variable masks the drastic difference between audio-only and audio-visual visits regarding the capture of SDoH Z codes. The use of audio-visual visits for low-risk, routine follow-up of pregnant women with nursing/midwife professionals, and in which physical exams or laboratory testing are not paramount, could help better capture, albeit imperfectly, SDoH measures relative to nursing/midwife in-person outpatient/office visits.

Appendix

Table 5.

Appendix Table: International Classification of Diseases-10-Clinical Modification (ICD-10-CM) Diagnosis codes used to identify the comorbid conditions

Conditions	Codes
Asthma	J45.xx
Chronic renal disease	N03.xx, N04.xx, N05.xx, N18.xx, N25.xx, N26.xx, O26.83
Ischemic heart disease	I20.xx, I25.xx
Pre-existing diabetes	E10.xx, E11.xx
Pre-existing hypertension	I10.xx, I11.xx, I12.xx, I13.xx, I15.xx, O10.xx
Obesity	O26.xx, O99.21, E66.xx
Alcohol use disorder	F10.xx
Tobacco use disorder	Z72.0x, O99.33, F17.xx
Other substance use disorder	F10.xx, F11.xx, F12.xx, F13.xx, F14.xx, F15.xx, F16.xx, F18.xx, F19.xx
Gestational diabetes	O24.4x, Z86.32
Gestational hypertension	O13.xx
Multiple gestation	Z37.2x, Z37.3x, Z37.7x
Previous Cesarean delivery	O34.2x
Mild Pre-eclampsia	O14.0x, O14.9x
Severe Pre-eclampsia	O14.1x

Authors’ contributions

MA conceptualized and designed the study, performed data analysis, wrote the initial draft, and reviewed the final draft for submission. NBN contributed to data validation, and critically reviewed and approved the final draft. CAB and MMA contributed to the study design, and critically reviewed and approved the final draft. HE contributed to the study conceptualization and study design. Unfortunately, he passed away before the final draft and could not review and approve the final draft.

Funding

This publication was made possible by grant number U3GRH40001 from the Office for the Advancement of Telehealth, Health Resources and Services Administration, DHHS. The information, conclusions, and opinions expressed are those of the authors, and no endorsement by OAT, HRSA, or DHHS is intended or should be inferred.

Data availability

The data analyzed in this study are proprietary and are only available through the Epic Cosmos data science server to the authorized users.

Declarations

Ethics approval and consent to participate

The study was determined as Not Human Subject Research by the Institutional Review Board, University of Arkansas for Medical Sciences under IRB #297774.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.