Determinants of Complementary and Integrative Health Approaches: A Cross-Sectional Study Using the All of Us Research Program

Introduction

The term “complementary and integrative health” (CIH) incorporates a broad spectrum of health care systems, practices, and products that are not typically included in conventional medicine. ¹ These approaches are often categorized based on their primary therapeutic approach, including nutritional, psychological, physical, and integrative approaches. ² Data from the National Health Interview Survey (NHIS) reveal that 42% of U.S. adults reported using at least one CIH approach in the past year. ³

Current trends in clinical practice guidelines (CPGs) and scholarly publications demonstrate increased attention given to CIH approaches. A 2017 systematic review on chronic pain management highlighted the frequent endorsement of nonpharmacological approaches in CPGs. Among 12 included CPGs, 2 recommended complementary therapies, 2 recommended dietary modification, 6 recommended psychologically based therapies, and 8 recommended physical therapy. ⁴ Furthermore, CIH-related scholarly publications have increased substantially since 2000. However, evaluating the prevalence of CIH utilization has primarily relied on population surveys, as constraints in health insurance coverage for CIH impede the collection of administrative data. ⁵

The National Institutes of Health’s All of Us (AoU) research program has enrolled over 800,000 participants, with recruiting efforts focusing on individuals from diverse backgrounds. More than 13,000 ongoing research projects and 400 publications have utilized AoU datasets. However, fewer than 1% of these projects, and none of the publications, have investigated CIH approaches. Given the increasing inclusion of CIH interventions in CPGs and the expanding body of CIH literature, our objective was to estimate the prevalence and predictors of CIH utilization within the AoU electronic health record (EHR) dataset.

Methods

We conducted a retrospective, cross-sectional analysis of AoU data to identify CIH users. To ensure data consistency, AoU utilizes the Observational Medical Outcomes Partnership (OMOP) Common Data Model. This model includes standardized vocabularies that map various coding systems into a unified format, allowing for consistent data representation across different health care datasets. Athena, a web-based application, supports the exploration of standardized OMOP terminology, allowing researchers to identify specific procedures from EHR data. Relevant procedural codes associated with CIH approaches, as defined by the National Center for Complementary and Integrative Health, ² were identified using the Athena database. We used these codes to systematically identify CIH users within the EHR dataset.

The Researcher Workbench enables access to deidentified AoU data for registered researchers, allowing workspace creation, cohort building, and data analysis (e.g., Jupyter Notebooks). The program gathers data from surveys, EHR, physical measurements, genomics, and wearable devices. Demographic variables, self-rated health status (via PROMIS Global Health), and lifestyle habits are available via three required surveys completed at baseline. Participants receiving care at any of 62 participating health care systems can opt to share EHR data, covering four domains: Conditions and Procedures, Drugs and Medications, Measurements, and Visits.

Results

A comprehensive search of the Athena database revealed 11 distinct CIH approaches (Supplementary Data S1). Five specific CIH approaches were identified within the AoU EHR dataset, including chiropractic manipulative therapy (CMT), osteopathic manipulative treatment (OMT), acupuncture, dietary modification, and massage therapy (Supplementary Data S2). Among participants who completed the initial AoU enrollment criteria (n = 410,264 as of September 2024), 5125 were identified as CIH users, and 202,864 were classified as nonusers. Of the CIH users, 3171 had complete survey data and were included in the analysis (Table 1).

Large SMD values (>0.3) ⁶ were observed for education and race, indicating that CIH users were more likely to be White and have some college education and less likely to be Black compared to nonusers. Moderate between-group differences (SMD 0.1–0.3) ⁶ were identified for age (older), education (college graduate), employment (not employed), marital status (married), insurance (not Medicaid), race (Other), and self-rated health status (lower) when comparing CIH users with nonusers.

The GLMs identified several significant associations between participant characteristics and CIH utilization (Table 2). Higher education levels were generally associated with greater CIH use, except for CMT, where participants with less than a high school education had higher odds compared to college graduates (OR: 4.09, 95% CI: 2.33–7.19).

Self-rated physical health was inversely related to dietary modification; each one-point increase in PROMIS physical health T-score reduced the odds of dietary modification by 7%. Insurance type also influenced utilization patterns. Compared with employer-based insurance, Medicaid-insured participants were more likely to use CMT, OMT, and dietary modification but less likely to use acupuncture and massage. Those with other insurance types (e.g., Veterans) were more likely to use CMT and acupuncture but less likely to use other modalities.

Females had nearly twice the odds of utilizing OMT and dietary modification compared to males. Black race predicted higher CMT use (62% higher odds), while Asian race predicted greater acupuncture use (66% higher odds) relative to White participants.

Discussion

Given the absence of comparable studies examining the prevalence and predictors of CIH approaches using EHR data, our results are compared with population-based survey data and findings from a scoping review. This study highlights both the similarities and differences that emerge from these comparative evaluations.

Rodgers-Melnick et al. ⁷ utilized NHIS data to evaluate characteristics associated with nonpharmacological pain management (i.e., chiropractic, yoga/Tai Chi, massage, or meditation/guided imagery). Users were younger, more often female, non-Hispanic, and had higher education and income than nonusers. These associations remained significant after covariate adjustment. Notably, this study did not evaluate individual modalities. These findings align with our population characteristics as documented in Table 1.

We identified lower educational attainment to be associated with higher odds of CMT utilization, which is consistent with findings from a recent scoping review, ⁸ but in contrast with NHIS data. ⁷ Higher educational status was also associated with the other four CIH modalities, which aligns with NHIS survey data. ^7,9,10

Black race was positively associated with CMT utilization, which contrasts with NHIS data which found the opposite association. ¹¹ This discrepancy may be attributed to the higher prevalence of Black participants in the AoU dataset (20.4%) versus a 10-year sample of NHIS survey data (11.8%). ¹² Additionally, Asian race was a predictor of acupuncture use, a finding supported by previous studies using NHIS data. ¹³

Female sex demonstrated lower odds of CMT utilization but higher odds of OMT and dietary modification. The former finding contradicts NHIS data ⁷ and a scoping review ¹⁴ that identified a predominance of female chiropractic users, while the latter is consistent with NHIS data indicating higher odds of dietary modification among females. ⁹

To date, population-based surveys have served as the primary method for determining the epidemiology of CIH utilization. Well-documented concerns regarding survey data include recall bias, self-reporting bias, and underrepresentation of populations that are difficult to reach. Furthermore, surveys often provide limited granularity, capturing broad categories of CIH practices. Integrating survey data with clinical data offers an opportunity to improve and clarify our understanding of the demographic patterns associated with CIH utilization. The integration of CIH approaches into health care delivery systems, and subsequently into EHR data, would provide a more accurate reflection of population-based CIH utilization patterns.

Limitations

Due to the ethnic and racial diversity within the AoU dataset, our results may not be directly comparable to other population-based studies. However, they likely offer a more accurate reflection of the overall U.S. population. EHR data are sourced from participants receiving care within specific hospital systems and may not be generalizable to other populations. Additionally, given that CIH practitioners frequently operate outside of such hospital systems, the relatively low CIH procedure counts are unsurprising.

Conclusion

This study shows that CIH utilization varied by sociodemographic group, with higher education and White race distinguishing users from nonusers. Education, insurance, sex, race, marital status, and self-rated health were key factors associated with specific CIH modalities. The integration of AoU EHR data provides a more detailed understanding of CIH utilization. Further research is needed to address data gaps, especially for CIH approaches performed outside conventional health care settings.