Introduction
Social determinants of health (SDOH) defined at the community level and at the individual level have recently received increased attention [1]. “Social determinants of health are the nonmedical factors that influence health outcomes. They are the conditions in which people are born, grow, work, live, and age, and the wider set of forces and systems shaping the conditions of daily life. These forces and systems include economic policies and systems, development agendas, social norms, social policies, racism, climate change, and political systems” [2]. There are both individual determinants of risk – biology and behavior – and group-level determinants, such as neighborhood and community [Reference Diez-Roux3]. Both have salience for health risks, health status, and health outcomes [Reference Diez-Roux3]. Throughout all analyses of SDOH, racial and ethnic disparities have played a critical role [Reference Yao and Robert4]. Some have argued that community data can be used as a proxy for the time-consuming process of collecting individual-level data on SDOH [Reference Liaw, Krist and Tong5,Reference Bazemore, Cottrell and Gold6]. However, the risk of assuming the community context determines individual risk (ecological fallacy) or that the individual determines community risk (atomistic fallacy) are both concerns; they may be correlated, but they are distinct [Reference Diez-Roux3].
In this paper, SDOH measured at an individual level will be compared to population-level findings in patients with multiple chronic diseases (MCDs) enrolled in the Tipping Points project, which is a cluster randomized controlled trial (cRCT, PCORI grant # IHS-2017C3-8923) examining clinical and patient-centered outcomes. Multiple chronic conditions are assessed from electronic health records (EHRs) using the enhanced CCI, a weighted measure of prognostically important chronic diseases. Patients with a CCI ≥ 4 are at high risk for destabilization leading to unplanned hospital admission and/or increased disability [Reference Charlson, Charlson, Peterson, Marinopoulos, Briggs and Hollenberg7–Reference Charlson, Wells, Ullman, King, Shmukler and Catapano9]. Patients for this cRCT were recruited from 16 Federally Qualified Health Centers (FQHCs) that are designated Patient-Centered Medical Homes (PCMHs) in four health systems in New York City and Chicago that are part of Clinical Directors Network’s (CDN) or AllianceChicago’s practice-based research networks (PBRNs). The patients who received care from PCMHs (usual care group) were compared to an intervention group that received care at PCMHs with health coaches added to help patients focus on setting life goals, learning self-management, using positive affect and self-affirmation, and overcoming obstacles and stresses.
The objective of this analysis is to compare SDOH measured at both the individual and population levels in patients with high comorbidity enrolled in this RCT who received their primary care at PCMH-designated FQHCs in New York City and Chicago.
Materials and methods
Site and patient recruitment
Patients with multiple chronic conditions defined by having a Charlson Comorbidity Index ≥ 4 were cluster randomized from 16 FQHCs (8 in Chicago and 8 in New York) from four FQHC networks that were invited and agreed to participate in the trial. These FQHCs are all accredited Patient-Centered Medical Homes (PCMHs), which serve mostly low-income Black and Latino patients. In PCMHs, staff work to make sure patients receive the right care at the right time. The PCMH provides individualized care with multidisciplinary care teams [10,11]. The current analysis includes 1488 participants from four health systems: Health System 1 (n = 374), Health System 2 (n = 330), Health System 3 (n = 395), and Health System 4 (n = 389). This trial was reviewed and approved by Institutional Review Boards (IRBs) including BRANY, Clinical Directors Network (CDN), Weill Cornell, and Chicago Area IRB (CHAIRb) and registered with ClinicalTrials.gov (ID NCT04176510) [12].
Assessment of comorbidity
Cited in more than 43,000 publications, the CCI, the most extensively validated measure of the prognostic impact of multiple chronic illnesses, is a weighted measure of the burden of chronic disease that predicts long-term prognosis [Reference Charlson, Pompei, Ales and MacKenzie8]. Different weights are assigned for specific conditions and summed to find the score [Reference Charlson, Pompei, Ales and MacKenzie8]. It has been adapted to predict future cost [Reference Charlson, Wells, Ullman, King, Shmukler and Catapano9,Reference Charlson, Charlson, Peterson, Marinopoulos, Briggs and Hollenberg7], and a partial index that is adjusted for the exclusions in the Tipping Points RCT was used. The CCI can be assessed prospectively by a 5–10 minute interview, as well as through claims data or EHR. Thus, the CCI-based method to identify patients with MCDs can be embedded within any EHR system (see Appendix 1 – partial adjusted CCI score).
All established patients of the participating FQHCs with a CCI ≥ 4 were evaluated for eligibility. The CCI was calculated from EHR data and then verified by patients prior to enrollment. Health coaches asked participants questions pertaining to eligibility and exclusion criteria. Consent was obtained in writing when health coaches were in-person at the FQHCs and orally when patients were consented remotely. Documentation of consent was recorded in the study database management system.
Assessment of SDOH
SDOH data for Tipping Points patients are collected from several sources: individual-level data from baseline questionnaires from consented patients, FQHC EHR systems, and neighborhood-level data from publicly available neighborhood-level information linked to the patient’s current home address and zip code.
Individual-level SDOH assessment
Individual determinants are collected from patients through baseline questionnaires as well as directly from EHRs. At baseline, all participants complete a standardized questionnaire related to unmet SDOH needs based on the National Association of Community Health Center’s “Protocol for Responding to and Assessing Patients’ Assets, Risks, and Experiences” (PRAPARE), which is available in English and Spanish [13–15]. Patient’s insurance, education, housing, transportation, food insecurity, repeated and utility needs are recorded. SDOH information is also extracted from patients’ EHRs and then verified by the patients, including race and ethnicity. Only insurance data were primarily extracted from the EHR. All the other data, including race, ethnicity, education, marital status, living situation, employment, and specific challenges such as food, housing, and transportation, were obtained from the patient interview at baseline.
While some health systems have dedicated sections on their EHR for assessing SDOH, the four health systems each use different EHR platforms. Further, only some of the health systems specifically ask questions about SDOH needs, such as food insecurity or housing instability. There was no universal SDOH screening used across all four health systems.
Population-level SDOH assessment
Neighborhood-level determinants were assessed using patients’ home addresses geocoded and linked to publicly available data. The Area Deprivation Index (ADI) and the Social Deprivation Index (SDI) are composite neighborhood-level measures used to assess socioeconomic disadvantage or deprivation, but they differ in focus, methodology, and application. The ADI is designed to assess deprivation in geographic areas using a single number, whereas the SDI can be separated into its individual components.
Computation of ADI and SDI
We downloaded the lists of consented participants and removed identifiers other than addresses. We used the Decentralized Geomarker Assessment for Multi-Site Studies (DeGAUSS), which is a “decentralized method for geocoding and deriving community and individual-level environmental characteristics while maintaining the privacy of protected health information” [16,Reference DeGAUSS17] to match each home address to a census tract and census block. We then matched the geocodes with two datasets to determine SDI and ADI. SDI was determined by matching the patient’s census tract with raw measures and scores in the SDI [18]. SDI includes seven community characteristics from census tracts about poverty, education, housing, single parents, employment, and car ownership and is one of the most commonly employed measures of population-level social risk with individually identifiable components [18]. (See Appendix B) ADI was ascertained by matching the patient’s census block with ADI National and State ranks [Reference Knighton, Savitz, Belnap, Stephenson and VanDerslice19]. ADI uses 17 indicators from the census: 8 focused on poverty, 5 on housing, and 4 on employment [Reference Knighton, Savitz, Belnap, Stephenson and VanDerslice19]. (See Appendix C) The 2020 year data were used; there was no difference between different years for enrolled patients.
The ADI measure is constructed by ranking the ADI score from low to high for the USA by each 1 percent range of the ADI score. For both SDI and national ADI, scores range from 1–100, with 1, least disadvantaged, to 100, most disadvantaged in the U.S. [Reference Knighton, Savitz, Belnap, Stephenson and VanDerslice19]. State ADI ranks in deciles from a scale of 1, least disadvantaged, to 10, most disadvantaged in that state [Reference Kind and Buckingham20].
Statistical methods
The present analyses evaluate and compare aspects of individual and population-level SDOH. We employed the following statistical tests and techniques while accounting for potential site-level dependence, where applicable. The chi-square test was employed to assess the association between categorical variables. Kruskal–Wallis tests, a nonparametric method, were applied to compare the distribution of continuous variables across the levels of a categorical variable. Multiple regression analysis was conducted to explore the relationship between various predictor variables and the population-level SDOH index. Regression models with CCI as a dependent variable use a negative binomial generalized linear model with a lower tail truncation at 3 to account for CCI’s count nature and threshold within the sample. Since the data exhibited site-level dependence or clustering, we employed robust standard errors. These cluster-robust standard errors corrected the correlation between observations within the same FQHC, providing accurate standard errors and confidence intervals for the regression coefficients. Maximum likelihood factor analysis was applied in order to uncover an underlying single latent factor that explains the correlations among observed SDOH variables. A significance level (two-tailed alpha) of 0.05 was selected for all statistical tests. We conducted all statistical analyses using the software package Stata 17 with the appropriate libraries and functions for each test.
Results
Population-level social determinants of health
Figure 1 shows the 4 health systems and their participating 16 FQHCs according to both ADI and SDI. Since the ADI is heterogeneous between the Chicago and New York FQHCs and it consists of a single number that cannot be split into interpretable components, subsequent analysis will focus on the SDI, which is homogeneous across the health systems.
Individual-level SDOH measures
Table 1 presents patient characteristics by SDI tertiles. Patients were an average age of 59 years old; 73% were female, 42% were Latino or Hispanic, and 44% were Black/ African American. 41% had an education < 12 years and 67% were unemployed. All had a comorbidity score ≥ 4 (eligibility criteria): 49% had a score of 4–5, 35% had a score of 6–8, and 16% had a score of 9 or more. Overall, 22% of patients had moderate to severe depression. Those living in the highest SDI areas had lower education, more with comorbidity > 9 and were less likely to be Latino or white.
Table 2 describes individual-level social determinants evaluated among these 1488 patients. While only 5% reported not having housing, almost 26% were worried about losing their housing. About 10% experienced issues with obtaining food and with paying for their utilities. Additionally, 14% needed help with getting transportation to medical appointments. Patients with higher SDI had more difficulty getting utilities (p = 0.03) and medical transportation (p = 0.04). A Kruskal–Wallis test for heterogeneity shows the above variables had differences by health system and FQHC site, except for worries about losing housing (p < 0.05).
Comparing individual- and population-level SDOH measures
Assessments of individual- versus population-level SDOH measures from the SDI were then compared. Table 3 shows the measures covering certain domains which are quite different.
SDI = Social Deprivation Index.
Table 4 compares the domain measures of individual- versus population-level SDOH, and while the components are generally related, most do differ. The only exception is education of less than 12 years which is employed in both assessments. Of note, with an identical measure, 42% of individuals had an education < 12 years, but only 21% of those in the participants’ community had an education of less than 12 years, a significant difference (p < 0.001). Transportation also differs: 15% on the individual level have an issue with medical transportation, and 40% in communities do not have vehicles. While the numbers differ, the assessments are also qualitatively different. Housing does not differ, but the assessments, unhoused versus crowding differ qualitatively, as does the proxy measure of income – inability to get food versus poverty. Population- or area-level SDOH correlate with overall health in a community but do not reliably predict individual patient health outcomes. Thus, larger differences between the individual- and population-level measures of SDOH are largely related to education and transportation issues, but the qualitative differences in transportation are substantial.
SDI = Social Deprivation Index.
Individual SDOH measures as predictors of population SDOH
Table 5 presents multiple regression analyses that explore the relationship between the individual measures and the population-level SDOH. Since our data exhibited site-level dependence or clustering, we employed robust standard errors for our regression coefficients to obtain accurate standard errors. Table 5 Column (1) examines the significance of the individual SDOH measures from Table 2 in predicting the population SDI. Significant predictors of the population SDI from individual determinants of health are being unhoused, unable to get utilities, and lacking medical transportation. Table 5 Column (2) examines the significance of the individual significant SDOH measures from Column (1). Each of these individual SDOH measures also predicts population SDI individually.
Controls for age, race (Hispanic/Latino, Black), and CCI score. Robust standard errors adjusted for site-level dependence in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. CCI = Enhanced Charlson Comorbidity Index; SDI = Social Deprivation Index; SDOH = social determinants of health.
To construct a univariate (latent) composite individual SDOH index, we developed a maximum likelihood one-factor model for the predictor variables in Table 5. A single factor that explains over 90% of the variability among the individual SDOH measures is a composite measure of “worry about losing housing,” “inability to get utilities and clothing,” and “lack of medical transportation.” However, from Column (3) of Table 5, this composite individual SDOH index is not significant as a single predictor of the population-level SDI. Thus, the analysis confirms that individual-level SDOH are quite distinct from population-level measures.
While high comorbidity patients have significant individual-level SDOH challenges, individual SDOH does not relate significantly to comorbidity scores. On the other hand, fitting a truncated negative binomial generalized linear model with the composite, CCI score is predicted substantially by the constructed composite individual SDOH index as a dependent variable (p < 0.0001) controlling for age and race (Hispanic/Latino, Black). Similarly, the CCI score is significantly predicted by the state-level ADI index (p < 0.0001).
Qualitative analysis
Unmet SDOH needs were discussed at our Community and Patient Stakeholder Advisory Committee (CPSAC) meetings in June 2023 and January 2024. Committee members include FQHC clinicians, staff, patients, and community members. In addition to the quantitative analysis, the CPSAC reviewed the data, shared their experiences and noted that a key issue with housing is the failure of landlords to make repairs. The CPSAC members identified the connections between the housing environment and mental health, highlighting the challenges associated with having landlords complete needed repairs.
As described in Figure 2, Advisory committee (CPSAC) members also stressed that medical transportation is an ongoing major problem. In addition to using mass transit, patients often book transportation through insurance or the Department of Transportation as part of Medicaid’s nonemergency medical transportation [21,22]. While patients can arrange low-cost or free medical transportation to take them to their medical appointments, the services are often unreliable, arriving late or not at all. Transportation appointments need to be scheduled at least 24 hours in advance, and often patients cannot reach the driver directly for updates. One stakeholder shared, “When they schedule an appointment and schedule the transportation for that appointment, the driver never arrives on time, and these clients are waiting for hours and hours.” When medical transportation delivers patients more than 15 minutes late to their appointments, the medical facilities often cancel the appointment as a policy. Patients are sent home to re-book their appointment, sometimes weeks later, and need to start the process again, thus delaying their access to care, and generating ongoing frustration: “…sometimes the doctor doesn’t know that they don’t come, they just put “no show” and they don’t come to the appointment because they had this problem with transportation…to schedule or reschedule an appointment again is a problem too because it takes 3 months or 6 months for a general doctor or for a specialist up to a year.”
Discussion
Socioeconomic status has always had salience for its explanatory power in health, but there is a contrast between individual-level and group-level determinants. Among these patients with high comorbidity enrolled in this cRCT conducted with 16 FQHCs in New York City and Chicago, the individual-level SDOH had minimal overlap with community-level SDOH.
The National Academy of Medicine [Reference Adler and Stead23,Reference Prather, Gottlieb and Giuse24] and the Centers for Medicare and Medicaid Services [25] have each recommended elements of SDOH to assess as part of care.
The American College of Physicians developed a comprehensive policy paper focused on addressing social determinants to improve patient care and health equity [Reference Daniel, Bornstein and Kane26]. It also stressed the importance of documenting individual-level impacts of SDOH [Reference Daniel, Bornstein and Kane26]. It emphasized the importance of expanding policy programs to reduce the socioeconomic inequalities with a negative impact on health and for investments in programs to reduce the disparities [Reference Daniel, Bornstein and Kane26].
Individual measures
The Institute of Medicine (IOM) also urged the incorporation of individual patient data on SDOH in EHRs [27]. The IOM defined SDOH as race, ethnicity, education, financial resources strain, connections and social isolation, and exposure to violence, as well as stress, depression, physical activity, tobacco, and alcohol use [27]. The 22 questions to assess these social and behavioral domains are clear and straightforward [Reference Adler and Stead23]. While the completion time is only 5 minutes, the questions have not been generally added to the EHR except for the Patient Health Questionnaire (PHQ-2) measure of depression [Reference Giuse, Koonce and Kusnoor28]. As noted previously, one standardized instrument adopted by many FQHCs is the PRAPARE, a 21-item survey that measures food insecurity, housing instability, financial resources strain, relationship safety, inadequate physical activity, social connection/isolation, and stress [14,Reference Giron, Cole and Nguyen29]. Some health systems have developed a modified version of PRAPARE to streamline the assessment of social needs [13]. A recent analysis of SDOH data collection in PCORnet Clinical Research Networks found that most health systems do not use a framework or standard terminologies for SDOH data [Reference Dullabh, Hovey and Leaphart30]. Moreover, 40% of health systems report a low percentage of patients with SDOH data [Reference Dullabh, Hovey and Leaphart30]. The PCORnet sites did agree on SDOH priority domains such as housing instability, food insecurity, transportation access, financial hardship, employment status, social isolation, intimate partner violence, and veteran status [Reference Dullabh, Hovey and Leaphart30]. One national study that examined 2333 physician practices and 757 hospitals reported that nearly 40% of physician practices and almost 80% of hospitals reported that they assessed transportation needs [Reference Fraze, Brewster, Lewis, Beidler, Murray and Colla31]. A more recent report examined 2749 hospital systems’ assessment of and intervention on unmet SDOH needs, including food insecurity, housing instability, utility needs, interpersonal violence, and transportation needs, and reported that transportation needs were the most commonly identified and intervened issue, in 77% of hospitals [Reference Sandhu, Liu, Gottlieb and Pantell32].
Community measures
The importance of area-level measures has been reinforced by recent analyses of county level cause-specific mortality rates from 1980 to 2014, which shows geographic disparities in life expectancy and rates of cause-specific deaths [Reference Dwyer-Lindgren, Bertozzi-Villa and Stubbs33,Reference Dwyer-Lindgren, Bertozzi-Villa and Stubbs34]. The geographic disparities were driven by race/ethnicity, socioeconomic status, and healthcare factors [Reference Dwyer-Lindgren, Bertozzi-Villa and Stubbs34]. Between the most advantaged and disadvantaged, there is a 15-year difference in life expectancy [Reference Chetty, Stepner and Abraham35]. Community-level social risks have been defined by a number of methods [Reference Bullard36]. A comprehensive analyses of socioeconomic gradients in health and area-based socioeconomic measures focused on geocoded data from two states [Reference Krieger, Chen, Waterman, Rehkopf and Subramanian37]. Evaluating census tracts with 4000, block groups with 1000, and zip codes with 30,000 people, they found that census tract analysis was the optimal strategy for geocoding for health [Reference Krieger, Chen, Waterman, Rehkopf and Subramanian37]. Evaluating mortality, low birth rate, cancer, and tuberculosis among other conditions, they found that the percentage below poverty was the most sensitive marker of socioeconomic gradients in health [Reference Krieger, Chen, Waterman, Rehkopf and Subramanian37]. Area-level measures have been increasingly seen as a method of accounting for social risk in healthcare payments.
There are many different area-level deprivation measures that are used [Reference Breslau, Martin, Timbie, Qureshi and Zajdman38]; two of the most commonly referenced are the SDI and the ADI. The SDI includes seven community characteristics from census tracts about the percentage living in poverty, less than high school education, rental housing, overcrowded (people vs. rooms) housing, single parents, unemployment, and having a car and is one of the most commonly employed measures of population-level social risk with individually identifiable components [18]. It was validated in relation to a measure of poverty [Reference Butler, Petterson, Phillips and Bazemore39]. Correlations between SDI and health outcomes (such as mortality, infant mortality, low birth rate, and diabetes) were lower because the units of geography differed for SDI (primary care service area) and health outcome (county) [Reference Butler, Petterson, Phillips and Bazemore39]. The goal is to use higher ADI levels to identify high-risk patients for care management interventions [Reference Knighton, Savitz, Belnap, Stephenson and VanDerslice19]. One analysis that focused on an area of relatively high poverty suggested that 30-day readmission was higher in higher ADI neighborhoods [Reference Hu, Kind and Nerenz40].
Individual versus population measures
A few years ago, the concept of hot spotting (focusing on the “super-utilizer” patients who were repeatedly in the ER and the hospital) was popularized by Gawande [Reference Gawande41]. This is still widely believed, although a RCT based on the Camden model found no difference in readmission after an intensive intervention with high utilizers [Reference Finkelstein, Zhou, Taubman and Doyle42]. Population segmentation is a misleading term, not related to a population, per se, but often referring to the use of prior utilization to stratify patients into tiers of utilization – which are not stable over time [Reference Johnson, Brewer and Estacio43].
However, the “hot spotting” of individuals led to the conceptualization of “cold spots” in communities where there are issues with housing, education, health, and broadly deficiencies in SDOH. In its most extreme construct, the focus should be on eliminating cold spots because they were responsible in large measure for patients’ healthcare problems and costs [Reference Westfall44]. While most analyses do not use that expansive a framework, communities as defined by census tracts have been characterized as “cold spots” – that is, those communities with worse incomes, education, and social deprivation [Reference Liaw, Krist and Tong5]. One study assessed 12 practices in one area and their census tracts according to the percentage of residents earning less than 200% of the poverty level, without high school diplomas, and according to the SDI [Reference Liaw, Krist and Tong5]. However, the differences in health measures such as cancer screening between whether patients live in a cold spot or not were modest except for differences in obesity [Reference Liaw, Krist and Tong5].
Yet the emphasis on community-level characteristics as defining individually relevant SDOH has persisted – in part because assessing communities from census tract or zip code is easier than collecting individual-level data. Several recent analyses contrasting individual- and community-level SDOH are comprehensive and persuasive about the distinctions between the two.
An important cross-sectional study of 36,578 patients at community health centers compared patient-level social risk data to community-level data through SDI [Reference Cottrell, Hendricks and Dambrun45]. Patient-level risks were assessed in those who completed at least one of three questions about financial strain, housing stability, and food insecurity, and the percentage of those who had one or more social risk factors was analyzed [Reference Cottrell, Hendricks and Dambrun45]. The individual risks were compared to quartiles of SDI, and social risks in the “coldest quartile” occurred in 30% of patients, similar to the second and third quartiles – 29% in each, while in the best off quartile, a total of 23% had social risks [Reference Cottrell, Hendricks and Dambrun45]. Thus overall, 30% of cold spot patients had social risks, and 28% of non-cold spot patients had social risks. Conversely, 69% of cold spot patients had no social risk, and 72% of non-cold spot patients had social risks [Reference Cottrell, Hendricks and Dambrun45].
Another analysis by the Society of Actuaries contrasted census tract data with SDOH data from 231,989 Medicaid patients [46]. Individual data was either from PRAPARE or from the assessment by the Accountable Health Communities Model [46,47]. Both surveys document education, housing, food insecurity, safety, transportation, income, and utilities. The analysis found that census-derived neighborhood social determinants were not significant predictors of individual SDOH for either adults or children [46]. The analysis then focused on both individual and neighborhood SDOH as predictors of utilization and cost and found that they differed in terms of who was predicted as high risk. They also found that individual and neighborhood SDOH were not always associated with inpatient cost, utilization, and high risk, and adding them to predictive models did not improve performance [46].
Limitations
The four health systems and 16 FQHC sites did not use a uniform method for assessing SDOH as part of their routine workflow. Additionally, the health systems used different EHR platforms. A final limitation is the reliability of address data as patients may have moved during the study. Patients provided their addresses at baseline and did not always update us if they moved.
Conclusion
In summary, patient- and community-level measures of SDOH provide different risk assessments. The use of community-level SDI data, while informative in the aggregate, when used to identify patients with individual unmet social needs, may be imprecise and could result in an ecologic fallacy [Reference Cottrell, Hendricks and Dambrun45]. The PCORnet Common Data Model contains the building blocks to calculate the comorbidity index, as well as various other indices including the ADI and SDI. ADI and SDI are calculated from raw zip codes and address data. The comorbidity index can be calculated from the ICD-10 codes in the PCORnet Common Data Model.
It is critical for health systems and practices to implement a standardized assessment of unmet SDOH needs that can be embedded within the EHR and workflow at entry into the practice and repeated at fixed intervals. This routine SDOH assessment needs to be combined with appropriate referrals within the practice and to external community-based partners that provide access to specific services. This study identified challenges at both the population and individual levels faced by patients with multiple chronic conditions that are associated with barriers to accessing primary care through medical transportation programs in two large urban settings. This systems-level barrier requires attention by health systems that are committed to improving access to care.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/cts.2024.598.
Acknowledgments
We wish to acknowledge the following FQHCs for their participation: Community Health Network (Crown Heights, Harlem, Long Island City, Sutphin), Family Health Centers (FHCs) at New York University (NYU) (Park Ridge, Park Slope, Sunset Park, Flatbush), and eight sites from two Chicago health networks. We would like to acknowledge the project team members and stakeholders from the participating sites: Nicholas Martin, Dr Taisha Benjamin, Ambrosia Elder, Dr Kelly Horn, Michael Chapman, Dr Rishi Dalal, Jodyann Wynter, and Dr Andrea Ciano (CHN) and Dr Radhika Gore, Phil Hayward, Dr Jorge Sastre, Kalisha Small and Rebecca Green, Claudya Verdiner, Dr Ekaterina Olkhina, Anne Marie Sabella, Dr Ramiro Jervis, and Dr Sandeep Bhat (FHCs at NYU).
Thank you to the Tipping Points research team, both current and former members: Mirta Milanes, Shelly Sital, Dr Nivedita Mohanty, Dr Eve Walter, Ariana Perdomo, Robin Andrews and Lewis Perin, Harpreet Kaur, Eilyn Candelo, Jacqueline Cruz, Maria Ruiz, Martha Muñoz, Victoria Sarita, Anisa Mian, Roxane Padilla, Andy Cruz, Jaileen Ocasio, Stephanie Vasquez, Andrea Chavarria, Azalia Garcia, Linda Humaidan, Joel Blanco-Aguirre, Nataly Aguirre, Edward Castillo, Annette Rueda, Nees Calderon, and Cynthia Mofunanya.
Thank you to the Community and Patient Stakeholder Advisory Committee and to our data partners at INSIGHT (Dr Mark Weiner, Alexandra LaMar, Catherine Rabin, Rosie Ferris, Kanta Hague, Dmitry Morozyuk, Joshua Gelber, Peter Morrisey), CAPriCORN (Shelly Sital, Andrea VanderLaan, Kyra VanDoren), Healthix (Todd Rogow, Tim Tirrell, and Tom Moore), and BronxRHIO (Kathryn Miller, Megha Khatri Arora, Ralph Figueroa, and Jianwen Wu).
Author contributions
Conceptualization, MEC, MTW, and JNT; methodology, MEC, MTW, JH, AC, and JNT; formal analysis, MEC, MTW, JH, RR, AC, TL, and JNT; resources, MEC, AC, and JNT; data curation, MEC, MTW, JH, RR, MG, AE, AC, TL, IM, and JNT; original draft preparation, MEC, MTW, RR, AC, IM, and JNT; writing—review and editing, MEC, MTW, JH, RR, GMM, MG, AC, TL, IM, AE, and JNT; visualization, MEC, MTW, JH, and RR; supervision, MEC, RR, AC, IM, and JNT; project administration, MEC, RR, AC, and JNT; funding acquisition, MEC, MTW, AC, and JNT. All authors have read and agreed to the published version of the manuscript.
Disclaimer
All statements in this report, including its findings and conclusions, are solely those of the authors and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute (PCORI), its Board of Governors, or its Methodology Committee.
Funding statement
This work was supported through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (IHS-2017C3-8923 to CDN), with additional infrastructure support provided by the Agency for Healthcare Research and Quality (AHRQ N2-PBRN Grant #1 P30-HS-021667 to CDN), INSIGHT (PCORI® Award #R-1306-03961 to Weill Cornell Medicine), and CAPriCORN (PCORI® Award #CDRN-1306-04737 to Northwestern Medicine). PCORI scientific staff played no role in study design and conduct. The statements presented in this publication are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors, or Methodology Committee. Dr Tobin is a founding investigator and serves on the Governance Boards of both INSIGHT and CAPriCORN Patient-Centered Outcomes Research Network (PCORnet®) Clinical Research Networks.
Competing interests
MEC at Cornell University has filed a patent application on methods and systems implementing the enhanced CCI in the management of healthcare resources. Cornell owns the copyright on the index, and MEC receives a portion of license revenue from Cornell University under Cornell’s Inventions and Related Property Rights Policy. The enhanced CCI has been disclosed to, and is managed by, the Center for Technology Licensing at Cornell University (CTL), the technology transfer arm of Cornell University. CTL has filed for patent protection on methods and systems implementing the CCI and also owns and administers the copyright. The CCI is available for licensing through CTL. To date, the CCI has been made widely available and licensed to many other academic institutions and noncommercial entities on a nonexclusive and royalty-free basis for research purposes only and typically in the context of a collaboration. Licenses to commercial entities are also available and are subject to negotiation with CTL.