Embedded multilevel regression and poststratification: Model-based inference with incomplete auxiliary information.

Health disparity research often evaluates health outcomes across demographic subgroups. Multilevel regression and poststratification (MRP) is a popular approach for small subgroup estimation as it can stabilize estimates by fitting multilevel models and adjust for selection bias by poststratifying on auxiliary variables, which are population characteristics predictive of the analytic outcome. However, the granularity and quality of the estimates produced by MRP are limited by the availability of the auxiliary variables' joint distribution; data analysts often only have access to the marginal distributions. To overcome this limitation, we embed the estimation of population cell counts needed for poststratification into the MRP workflow: embedded MRP (EMRP). Under EMRP, we generate synthetic populations of the auxiliary variables before implementing MRP. All sources of estimation uncertainty are propagated with a fully Bayesian framework. Through simulation studies, we compare different methods of generating the synthetic populations and demonstrate EMRP's improvements over alternatives on the bias-variance tradeoff to yield valid subpopulation inferences of interest. We apply EMRP to the Longitudinal Survey of Wellbeing and estimate food insecurity prevalence among vulnerable groups in New York City. We find that all EMRP estimators can correct for the bias in classical MRP while maintaining lower standard errors and narrower confidence intervals than directly imputing with the weighted finite population Bayesian bootstrap (WFPBB) and design-based estimates. Performances from the EMRP estimators do not differ substantially from each other, though we would generally recommend using the WFPBB-MRP for its consistently high coverage rates.

Individual and Community level Developmental Adversities: Associations with Marijuana and Alcohol Use in Late-Adolescents and Young Adults.

Exposure to community and individual level stressors during adolescence has been reported to be associated with increased substance use. However, it remains unclear what the relative contribution of different community- and individual-level factors play when alcohol and marijuana use become more prevalent during late adolescence. The present study uses a large longitudinal sample of adolescents (Wave 1: N = 2017; 55% Female; 54.5% White, 22.3% Black, 8% Hispanic, 15% other) to evaluate the association and potential interactions between community- and individual-level factors and substance use from adolescence to young adulthood (Wave 1 to Wave 3 Age Mean [SD]: 16.7 [1.1], 18.3 [1.2], 19.3 [1.2]). Across three waves of data, multilevel modeling (MLM) is used to evaluate the association between community affluence and disadvantage, individual household socioeconomic status (SES, measured as parental level of education and self-reported public assistance) and self-reported childhood maltreatment with self-reported 12-month alcohol and 12-month marijuana use occasions. Sample-selection weights and attrition-adjusted weights are accounted for in the models to evaluate the robustness of the estimated effects. Across the MLMs, there is a significant positive association between community affluence and parental education with self-reported alcohol use but not self-reported marijuana use. In post hoc analyses, higher neighborhood affluence in older adolescents is associated with higher alcohol use and lower use in younger adolescents; the opposite association is found for neighborhood disadvantage. Consistent with past literature, there is a significant positive association between self-reported childhood maltreatment and self-reported 12-month alcohol and 12-month marijuana use. Results are largely consistent across weighted and unweighted analyses, however, in weighted analyses there is a significant negative association between community disadvantage and self-reported 12-month alcohol use. This study demonstrates a nuanced relationship between community- and individual-level factors and substance use during the transitional window of adolescence which should be considered when contextualizing and interpreting normative substance use during adolescence.

Beyond Vaccination Rates: A Synthetic Random Proxy Metric of Total SARS-CoV-2 Immunity Seroprevalence in the Community.

BACKGROUND: Explicit knowledge of total community-level immune seroprevalence is critical to developing policies to mitigate the social and clinical impact of SARS-CoV-2. Publicly available vaccination data are frequently cited as a proxy for population immunity, but this metric ignores the effects of naturally acquired immunity, which varies broadly throughout the country and world. Without broad or random sampling of the population, accurate measurement of persistent immunity post-natural infection is generally unavailable. METHODS: To enable tracking of both naturally acquired and vaccine-induced immunity, we set up a synthetic random proxy based on routine hospital testing for estimating total immunoglobulin G (IgG) prevalence in the sampled community. Our approach analyzed viral IgG testing data of asymptomatic patients who presented for elective procedures within a hospital system. We applied multilevel regression and poststratification to adjust for demographic and geographic discrepancies between the sample and the community population. We then applied state-based vaccination data to categorize immune status as driven by natural infection or by vaccine. RESULTS: We validated the model using verified clinical metrics of viral and symptomatic disease incidence to show the expected biologic correlation of these entities with the timing, rate, and magnitude of seroprevalence. In mid-July 2021, the estimated immunity level was 74% with the administered vaccination rate of 45% in the two counties. CONCLUSIONS: Our metric improves real-time understanding of immunity to COVID-19 as it evolves and the coordination of policy responses to the disease, toward an inexpensive and easily operational surveillance system that transcends the limits of vaccination datasets alone.

Routine Hospital-based SARS-CoV-2 Testing Outperforms State-based Data in Predicting Clinical Burden.

Throughout the coronavirus disease 2019 (COVID-19) pandemic, government policy and healthcare implementation responses have been guided by reported positivity rates and counts of positive cases in the community. The selection bias of these data calls into question their validity as measures of the actual viral incidence in the community and as predictors of clinical burden. In the absence of any successful public or academic campaign for comprehensive or random testing, we have developed a proxy method for synthetic random sampling, based on viral RNA testing of patients who present for elective procedures within a hospital system. We present here an approach under multilevel regression and poststratification to collecting and analyzing data on viral exposure among patients in a hospital system and performing statistical adjustment that has been made publicly available to estimate true viral incidence and trends in the community. We apply our approach to tracking viral behavior in a mixed urban-suburban-rural setting in Indiana. This method can be easily implemented in a wide variety of hospital settings. Finally, we provide evidence that this model predicts the clinical burden of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) earlier and more accurately than currently accepted metrics. See video abstract at, http://links.lww.com/EDE/B859.

Conceptualizing protective family context and its effect on substance use: Comparisons across diverse ethnic-racial youth.

Background: Although family behaviors are known to be important for buffering youth against substance use, research in this area often evaluates a particular type of family interaction and how it shapes adolescents' behaviors, when it is likely that youth experience the co-occurrence of multiple types of family behaviors that may be protective. Methods: The current study (N = 1716, 10th and 12th graders, 55% female) examined associations between protective family context, a latent variable comprised of five different measures of family behaviors, and past 12 months substance use: alcohol, cigarettes, marijuana, and e-cigarettes. Results: A multi-group measurement invariance assessment supported protective family context as a coherent latent construct with partial (metric) measurement invariance among Black, Latinx, and White youth. A multi-group path model indicated that protective family context was significantly associated with less substance use for all youth, but of varying magnitudes across ethnic-racial groups. Conclusion: These results emphasize the importance of evaluating psychometric properties of family-relevant latent variables on the basis of group membership in order to draw appropriate inferences on how such family variables relate to substance use among diverse samples.

A National Study of the Use of Asymptomatic Systemic Imaging for Surveillance Following Breast Cancer Treatment (AFT-01).

BACKGROUND: Although not guideline recommended, studies suggest 50% of locoregional breast cancer patients undergo systemic imaging during follow-up, prompting its inclusion as a Choosing Wisely measure of potential overuse. Most studies rely on administrative data that cannot delineate scan intent (prompted by signs/symptoms vs. asymptomatic surveillance). This is a critical gap as intent is the only way to distinguish overuse from appropriate care. OBJECTIVE: Our aim was to assess surveillance systemic imaging post-breast cancer treatment in a national sample accounting for scan intent. METHODS: A stage-stratified random sample of 10 women with stage II-III breast cancer in 2006-2007 was selected from each of 1217 Commission on Cancer-accredited facilities, for a total of 10,838 patients. Registrars abstracted scan type (computed tomography [CT], non-breast magnetic resonance imaging, bone scan, positron emission tomography/CT) and intent (cancer-related vs. not, asymptomatic surveillance vs. not) from medical records for 5 years post-diagnosis. Data were merged with each patient's corresponding National Cancer Database record, containing sociodemographic and tumor/treatment information. RESULTS: Of 10,838 women, 30% had one or more, and 12% had two or more, systemic surveillance scans during a 4-year follow-up period. Patients were more likely to receive surveillance imaging in the first follow-up year (lower proportions during subsequent years) and if they had estrogen receptor/progesterone receptor-negative tumors. CONCLUSIONS: Locoregional breast cancer patients undergo asymptomatic systemic imaging during follow-up despite guidelines recommending against it, but at lower rates than previously reported. Providers appear to use factors that confer increased recurrence risk to tailor decisions about systemic surveillance imaging, perhaps reflecting limitations of data on which current guidelines are based. ClinicalTrials.gov Identifier: NCT02171078.

Bayesian inference under cluster sampling with probability proportional to size.

Cluster sampling is common in survey practice, and the corresponding inference has been predominantly design based. We develop a Bayesian framework for cluster sampling and account for the design effect in the outcome modeling. We consider a two-stage cluster sampling design where the clusters are first selected with probability proportional to cluster size, and then units are randomly sampled inside selected clusters. Challenges arise when the sizes of the nonsampled cluster are unknown. We propose nonparametric and parametric Bayesian approaches for predicting the unknown cluster sizes, with this inference performed simultaneously with the model for survey outcome, with computation performed in the open-source Bayesian inference engine Stan. Simulation studies show that the integrated Bayesian approach outperforms classical methods with efficiency gains, especially under informative cluster sampling design with small number of selected clusters. We apply the method to the Fragile Families and Child Wellbeing study as an illustration of inference for complex health surveys.

Utility of Clinical Breast Examinations in Detecting Local-Regional Breast Events After Breast-Conservation in Women with a Personal History of High-Risk Breast Cancer.

BACKGROUND: Although breast cancer follow-up guidelines emphasize the importance of clinical examinations, prior studies suggest a small fraction of local-regional events occurring after breast conservation are detected by examination alone. Our objective was to examine how local-regional events are detected in a contemporary, national cohort of high-risk breast cancer survivors. METHODS: A stage-stratified sample of stage II/III breast cancer patients diagnosed in 2006-2007 (n = 11,099) were identified from 1217 facilities within the National Cancer Data Base. Additional data on local-regional and distant breast events, method of event detection, imaging received, and mortality were collected. We further limited the cohort to patients with breast conservation (n = 4854). Summary statistics describe local-regional event rates and detection method. RESULTS: Local-regional events were detected in 5.5 % (n = 265) of patients. Eighty-three percent were ipsilateral or contralateral in-breast events, and 17 % occurred within ipsilateral lymph nodes. Forty-eight percent of local-regional events were detected on asymptomatic breast imaging, 29 % by patients, and 10 % on clinical examination. Overall, 0.5 % of the 4854 patients had a local-regional event detected on examination. Examinations detected a higher proportion of lymph node events (8/45) compared with in-breast events (18/220). No factors were associated with method of event detection. DISCUSSION: Clinical examinations, as an adjunct to screening mammography, have a modest effect on local-regional event detection. This contradicts current belief that examinations are a critical adjunct to mammographic screening. These findings can help to streamline follow-up care, potentially improving follow-up efficiency and quality.

LATENT SUBGROUP IDENTIFICATION IN IMAGE-ON-SCALAR REGRESSION.

Image-on-scalar regression has been a popular approach to modeling the association between brain activities and scalar characteristics in neuroimaging research. The associations could be heterogeneous across individuals in the population, as indicated by recent large-scale neuroimaging studies, for example, the Adolescent Brain Cognitive Development (ABCD) Study. The ABCD data can inform our understanding of heterogeneous associations and how to leverage the heterogeneity and tailor interventions to increase the number of youths who benefit. It is of great interest to identify subgroups of individuals from the population such that: (1) within each subgroup the brain activities have homogeneous associations with the clinical measures; (2) across subgroups the associations are heterogeneous, and (3) the group allocation depends on individual characteristics. Existing image-on-scalar regression methods and clustering methods cannot directly achieve this goal. We propose a latent subgroup image-on-scalar regression model (LASIR) to analyze large-scale, multisite neuroimaging data with diverse sociode-mographics. LASIR introduces the latent subgroup for each individual and group-specific, spatially varying effects, with an efficient stochastic expectation maximization algorithm for inferences. We demonstrate that LASIR outperforms existing alternatives for subgroup identification of brain activation patterns with functional magnetic resonance imaging data via comprehensive simulations and applications to the ABCD study. We have released our reproducible codes for public use with the software package available on Github.

Advancing high quality longitudinal data collection: Implications for the HEALthy Brain and Child Development (HBCD) Study design and recruitment.

The HEALthy Brain and Child Development (HBCD) Study, a multi-site prospective longitudinal cohort study, will examine human brain, cognitive, behavioral, social, and emotional development beginning prenatally and planned through early childhood. The HBCD Study aims to reflect the sociodemographic diversity of pregnant individuals in the U.S. The study will also oversample individuals who use substances during pregnancy and enroll similar individuals who do not use to allow for generalizable inferences of the impact of prenatal substance use on trajectories of child development. Without probability sampling or a randomization-based design, the study requires innovation during enrollment, close monitoring of group differences, and rigorous evaluation of external and internal validity across the enrollment period. In this article, we discuss the HBCD Study recruitment and enrollment data collection processes and potential analytic strategies to account for sources of heterogeneity and potential bias. First, we introduce the adaptive design and enrollment monitoring indices to assess and enhance external and internal validity. Second, we describe the visit schedule for in-person and remote data collection where dyads are randomly assigned to visit windows based on a jittered design to optimize longitudinal trajectory estimation. Lastly, we provide an overview of analytic procedures planned for estimating trajectories.

Multiple Imputation with Massive Data: An Application to the Panel Study of Income Dynamics.

Multiple imputation (MI) is a popular and well-established method for handling missing data in multivariate data sets, but its practicality for use in massive and complex data sets has been questioned. One such data set is the Panel Study of Income Dynamics (PSID), a longstanding and extensive survey of household income and wealth in the United States. Missing data for this survey are currently handled using traditional hot deck methods because of the simple implementation; however, the univariate hot deck results in large random wealth fluctuations. MI is effective but faced with operational challenges. We use a sequential regression/chained-equation approach, using the software IVEware, to multiply impute cross-sectional wealth data in the 2013 PSID, and compare analyses of the resulting imputed data with those from the current hot deck approach. Practical difficulties, such as non-normally distributed variables, skip patterns, categorical variables with many levels, and multicollinearity, are described together with our approaches to overcoming them. We evaluate the imputation quality and validity with internal diagnostics and external benchmarking data. MI produces improvements over the existing hot deck approach by helping preserve correlation structures, such as the associations between PSID wealth components and the relationships between the household net worth and sociodemographic factors, and facilitates completed data analyses with general purposes. MI incorporates highly predictive covariates into imputation models and increases efficiency. We recommend the practical implementation of MI and expect greater gains when the fraction of missing information is large.

Association Between Organizational Quality and Out-of-Network Primary Care Among Accountable Care Organizations That Care for High vs Low Proportions of Patients of Racial and Ethnic Minority Groups.

Importance: Medicare accountable care organizations (ACOs) that disproportionately care for patients of racial and ethnic minority groups deliver lower quality care than those that do not, potentially owing to differences in out-of-network primary care among them. Objective: To examine how organizational quality is associated with out-of-network primary care among ACOs that care for high vs low proportions of patients of racial and ethnic minority groups. Design Setting and Participants: A retrospective cohort study was conducted between March 2019 and October 2021 using claims data (2013 to 2016) from a national sample of Medicare beneficiaries. Among beneficiaries who were assigned to 1 of 528 Medicare ACOs, a distinction was made between those treated by organizations that cared for high (vs low) proportions of patients of racial and ethnic minority groups. For each ACO, the amount of out-of-network primary care that it delivered annually was determined. Multivariable models were fit to evaluate how the quality of care that beneficiaries received varied by the proportion of care provided to patients of racial and ethnic minority groups by the ACO and its amount of out-of-network primary care. Exposures: The degree of care provided to patients of racial and ethnic minority groups by the ACO and its amount of out-of-network primary care. Main Outcomes and Measures: The ACO quality assessed with 5 preventive care services and 4 utilization metrics. Results: Among 3 955 951 beneficiary-years (2 320 429 [58.7%] women; 71 218 [1.8%] Asian, 267 684 [6.8%] Black, 44 059 [1.1%] Hispanic, 4922 [0.1%] North American Native, and 3 468 987 [87.7%] White individuals and 56 157 [1.4%] of Other race and ethnicity), those assigned to ACOs serving many patients of racial and ethnic minority groups at the mean level of out-of-network primary care were less likely than those assigned to ACOs serving fewer patients of racial and ethnic minority groups to receive diabetic retinal examinations (predicted probability, 49.4% [95%CI, 49.0%-49.7%] vs 51.6% [95% CI, 51.5%-51.8%]), glycated hemoglobin testing (predicted probability, 58.5% [95% CI, 58.2%-58.5%] vs 60.4% [95% CI, 60.3%-60.6%]), or low-density lipoprotein cholesterol testing (predicted probability, 85.2% [95% CI, 85.0%-85.5%] vs 86.0% [95% CI, 85.9%-86.1%]). They were also more likely to experience all-cause 30-day readmissions (predicted probability, 16.4% [95% CI, 16.1%-16.7%] vs 15.7% [95% CI, 15.6%-15.8%]). However, as the level of out-of-network primary care decreased, these gaps closed substantially, such that beneficiaries at ACOs that served many and fewer patients of racial and ethnic minority groups in the lowest percentile of out-of-network primary care received care of comparable quality. Conclusions and Relevance: This large cohort study found that quality performance among ACOs serving many patients of racial and ethnic minority groups was negatively associated with their level of out-of-network primary care.

An empirical evaluation of alternative approaches to adjusting for attrition when analyzing longitudinal survey data on young adults' substance use trajectories.

OBJECTIVES: Longitudinal survey data allow for the estimation of developmental trajectories of substance use from adolescence to young adulthood, but these estimates may be subject to attrition bias. Moreover, there is a lack of consensus regarding the most effective statistical methodology to adjust for sample selection and attrition bias when estimating these trajectories. Our objective is to develop specific recommendations regarding adjustment approaches for attrition in longitudinal surveys in practice. METHODS: Analyzing data from the national U.S. Monitoring the Future panel study following four cohorts of individuals from modal ages 18 to 29/30, we systematically compare alternative approaches to analyzing longitudinal data with a wide range of substance use outcomes, and examine the sensitivity of inferences regarding substance use prevalence and trajectories as a function of college attendance to the approach used. RESULTS: Our results show that analyzing all available observations in each wave, while simultaneously accounting for the correlations among repeated observations, sample selection, and attrition, is the most effective approach. The adjustment effects are pronounced in wave-specific descriptive estimates but generally modest in covariate-adjusted trajectory modeling. CONCLUSIONS: The adjustments can refine the precision, and, to some extent, the implications of our findings regarding young adult substance use trajectories.

Surveillance Imaging vs Symptomatic Recurrence Detection and Survival in Stage II-III Breast Cancer (AFT-01).

BACKGROUND: Guidelines for follow-up after locoregional breast cancer treatment recommend imaging for distant metastases only in the presence of patient signs and/or symptoms. However, guidelines have not been updated to reflect advances in imaging, systemic therapy, or the understanding of biological subtype. We assessed the association between mode of distant recurrence detection and survival. METHODS: In this observational study, a stage-stratified random sample of women with stage II-III breast cancer in 2006-2007 and followed through 2016 was selected, including up to 10 women from each of 1217 Commission on Cancer facilities (n = 10 076). The explanatory variable was mode of recurrence detection (asymptomatic imaging vs signs and/or symptoms). The outcome was time from initial cancer diagnosis to death. Registrars abstracted scan type, intent (cancer-related vs not, asymptomatic surveillance vs not), and recurrence. Data were merged with each patient's National Cancer Database record. RESULTS: Surveillance imaging detected 23.3% (284 of 1220) of distant recurrences (76.7%, 936 of 1220 by signs and/or symptoms). Based on propensity-weighted multivariable Cox proportional hazards models, patients with asymptomatic imaging compared with sign and/or symptom detected recurrences had a lower risk of death if estrogen receptor (ER) and progesterone receptor (PR) negative, HER2 negative (triple negative; hazard ratio [HR] = 0.73, 95% confidence interval [CI] = 0.54 to 0.99), or HER2 positive (HR = 0.51, 95% CI = 0.33 to 0.80). No association was observed for ER- or PR-positive, HER2-negative (HR = 1.14, 95% CI = 0.91 to 1.44) cancers. CONCLUSIONS: Recurrence detection by asymptomatic imaging compared with signs and/or symptoms was associated with lower risk of death for triple-negative and HER2-positive, but not ER- or PR-positive, HER2-negative cancers. A randomized trial is warranted to evaluate imaging surveillance for metastases results in these subgroups.

Who Returns? Understanding Varieties of Longitudinal Participation in MIDUS.

Objectives: This study describes a major effort to reinstate dropouts from the MIDUS longitudinal study and compare baseline characteristics among subgroups of participants to better understand predictors of retention, attrition, and reinstatement. Methods: All living dropouts were contacted, and 651 reinstated participants were interviewed in person (31.4% response rate). Age, gender, education, marital status, parental status, and physical and mental health were compared among the following groups: longitudinal sample, reinstated sample, those fielded for reinstatement who did not return, and those who dropped out at the 2nd or 3rd wave. Results: Multivariate analyses revealed that reinstated participants were younger, male, unmarried, and less educated and had children at baseline compared to longitudinal participants. Reinstatement was unsuccessful among those with poorer mental health at baseline compared to longitudinal participants. Discussion: This study informs reinstatement efforts, adjustment for attrition bias, and use of post-baseline data to examine aging consequents of early life vulnerability.

Comparison of Selective Versus Empiric Pharmacologic Preventative Therapy With Kidney Stone Recurrence.

OBJECTIVE: To assess the effectiveness of an empiric approach to metabolic stone prevention. METHODS: Using medical claims from a cohort of working age adults with kidney stone diagnoses (2008-2017), we identified the subset who were prescribed thiazides, alkali therapy, or allopurinol-collectively known as preventive pharmacologic therapy (PPT). We distinguished between those who had 24-hour urine testing prior to initiating PPT (selective therapy) from those without it (empiric therapy). We conducted a survival analysis for time to first recurrence for stone-related events, including ED visits, hospitalizations, and surgery, up to 2 years after initiating PPT. RESULTS: Of 10,125 patients identified, 2744 (27%) and 7381 (73%) received selective and empiric therapy, respectively. The overall frequency of any stone-related event was 11%, and this did not differ between the 2 groups on bivariate analysis (P = .29). After adjusting for sociodemographic factors, comorbidities, medication class, and adherence, there was no difference in the hazard of a stone-related event between the selective and empiric therapy groups (hazard ratio, 0.97; 95% confidence interval, 0.84-1.12). When considered individually, the frequency of ED visits, hospitalizations, and surgeries did not differ between groups. Greater adherence to PPT and older age were associated with a lower hazard of a stone-related event (both P < .05). CONCLUSION: Compared to empiric therapy, PPT guided by 24-hour urine testing, on average, is not associated with a lower hazard of a stone-related event. These results suggest a need to identify kidney stone patients who benefit from 24-hour urine testing.

Bayesian profiling multiple imputation for missing hemoglobin values in electronic health records.

Electronic health records (EHRs) are increasingly used for clinical and comparative effectiveness research, but suffer from missing data. Motivated by health services research on diabetes care, we seek to increase the quality of EHRs by focusing on missing values of longitudinal glycosylated hemoglobin (A1c), a key risk factor for diabetes complications and adverse events. Under the framework of multiple imputation (MI), we propose an individualized Bayesian latent profiling approach to capture A1c measurement trajectories subject to missingness. The proposed method is applied to EHRs of adult patients with diabetes in a large academic Midwestern health system between 2003 and 2013 and had Medicare A and B coverage. We combine MI inferences to evaluate the association of A1c levels with the incidence of acute adverse health events and examine patient heterogeneity across identified patient profiles. We investigate different missingness mechanisms and perform imputation diagnostics. Our approach is computationally efficient and fits flexible models that provide useful clinical insights.

Population Weighting in Statistical Analysis.

This Guide to Statistics and Methods provides an overview of weighted analyses of population-based surveys, which can help achieve statistically valid, representative population-based findings.

Geographic Variations in Physician Relationships Over Time: Implications for Care Coordination.

Care coordination may be more challenging when the specific physicians with whom primary care physicians (PCPs) are expected to coordinate care change over time. Using Medicare data on physician patient-sharing relationships and the Dartmouth Atlas, we explored the extent to which PCPs tend to share patients with other physicians over time. We found that 70.7% of ties between PCPs and other physicians that were present in 2012 persisted in 2013, and additional shared patients in 2012 increased the odds of being connected in 2013. Regions with higher persistent ties tended to have lower rates of emergency room visits, and regions where PCPs had more physician connections were more likely to have higher emergency room visits. The results point to potential opportunities and challenges faced by health care reforms that seek to improve coordination.