José R. Zubizarreta’s research while affiliated with Harvard Medical School and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (120)


Predictors of being in the subgroup with higher versus low expected compliance with guided i‐CBT. 12‐M, recall period in the past 12 months; 2‐WK, recall period in the past 2 weeks; 30‐D, recall period in the past 30 days; ADHD, attention‐deficit hyperactivity disorder; COVID‐19, severe acute respiratory syndrome coronavirus 2; i‐CBT, guided internet‐delivered cognitive behavioral therapy; MDE, major depressive episode; NSSI, non‐suicidal self‐injury; PHQ‐ADS, the 0–48 Patient Health Questionnaire Anxiety and Depression Scale; SAD, social anxiety disorder; SHAP, Shapley Additive Explanations.
 aThe SHAP values for a particular predictor may vary among participants with the same score on that predictor due to interactions with other predictors, resulting in variations in the sign of the relationship between the predictor and the outcome. The dominant direction of relationship was determined by visually examining the beeswarm plot displayed on the right side of the figure.
 bKey predictors are characterized as the top 5 predictors within each domain with mean absolute SHAP values of 5.0% or higher.
 cSum of 2‐W 9‐item 0–27 Patient Health Questionnaire score, 2‐W 7‐item 0–21 Generalized Anxiety Disorder score.
 dBased on the Leeds Risk Index (score range 0–16), where higher scores denote greater risk for poorer response to face‐to‐face psychotherapy (Delgadillo et al., 2016, 2020).
Statistical methods to adjust for the effects on intervention compliance in randomized clinical trials where precision treatment rules are being developed
  • Article
  • Full-text available

January 2025

·

54 Reads

·

·

·

[...]

·

Ronald C. Kessler

Background Heterogeneity of treatment effects (HTEs) can occur because of either differential treatment compliance or differential treatment effectiveness. This distinction is important, as it has action implications, but it is unclear how to distinguish these two possibilities statistically in precision treatment analysis given that compliance is not observed until after randomization. We review available statistical methods and illustrate a recommended method in secondary analysis in a trial focused on HTE. Methods The trial randomized n = 880 anxious and/or depressed university students to guided internet‐delivered cognitive behavioral therapy (i‐CBT) or treatment‐as‐usual (TAU) and evaluated joint remission. Previously reported analyses documented superiority of i‐CBT but significant HTE. In the reanalysis reported here, we used baseline (i.e., pre‐randomization) covariates to predict compliance among participants randomized to guided i‐CBT, generated a cross‐validated within‐person expected compliance score based on this model in both intervention groups, and then used this expected composite score as a predictor in an expanded HTE analysis. Results The significant intervention effect was limited to participants with high expected compliance. Residual HTE was nonsignificant. Conclusions Future psychotherapy HTE trials should routinely develop and include expected compliance composite scores to distinguish the effects of differential treatment compliance from the effects of differential treatment effectiveness.

Download

Targeting Operator Characteristic curve for ATE in the test sample (n = 1052). ATE, average treatment effect of exposure to high combat‐related stressors (CRS) on persistent post‐traumatic stress disorder (PTSD) defined as the difference in predicted probability of persistent PTSD in the presence versus absence of high CRS after adjusting for nonrandom exposure to high CRS with respect to the range of variables assessed in the baseline survey; Treated fraction (q), the proportion of soldiers who would either be prevented from being exposed to high CRS or protected from the effects of exposure by some to be determined intervention.
Key predictorsa (defined as those with SHAPP values of at least 5.0%) of differential resilience to the effects of high CRS on persistent post‐deployment PTSD defined by SHAP value analysis among combat arms soldiers in the STARRS PPDS sample (n = 2542)b. aKey predictors are defined as those with mean absolute SHAP values at least 5% as large as the mean absolute SHAP value of the total model. The latter is 2.1%, which means that predicted CATE for a given individual would have changed by an average of 2.1% if all significant predictors were set to the sample mean rather than to the observed values. bThe SHAP value analysis was carried out using the xgboost R package (Chen et al., 2024). The XGBoost algorithm (Chen & Guestrin, 2016) was used to predict individual differences in estimated CATE from the pre‐deployment predictors that were used in grf (Athey et al., 2024). to generate individual‐level estimates of CATE. Xgboost implements the tree SHAP method of estimating SHAP values (Lundberg & Lee, 2017). cThe SHAP value of a predictor can vary across respondents whenever there are nonlinearities and/or interactions in the model. This variation can be characterized in a bee swarm plot in which the SHAP value for each respondent is treated as a dot and values of the SHAP value are defined by the X Axis. A red dot means a higher value of the predictor. A blue dot means a lower value. If the red dots are predominantly on the right side of the swarm, it means that high scores on the predictor are associated with higher CATEs, which is equivalent to lower resilience. dAs noted above in FN a, a key predictor is defined as one with a mean absolute SHAP value at least 5% as large as the 2.1% mean absolute SHAP value of all predictors in the model. This means that any individual predictor with a mean absolute SHAP value of 0.01% (i.e., 5% of 2.1%) would be considered a key predictor. For ease of interpretation, we report proportional SHAP values (SHAPP) directly in the table. For example, a predictor with a mean absolute SHAP value of, say, 0.01% would be reported as having a SHAPP) of exactly 5.0% (i.e., 0.01/2.1). CRS, combat‐related stressors; PTSD, post‐traumatic stress disorder; SHAP, Shapley Additive Explanations; STARRS PPDS, Army Study to Assess Risk and Resilience in Servicemembers Pre‐Post Deployment Study.
A prediction model for differential resilience to the effects of combat‐related stressors in US army soldiers

October 2024

·

100 Reads

·

1 Citation

Objectives To develop a composite score for differential resilience to effects of combat‐related stressors (CRS) on persistent DSM‐IV post‐traumatic stress disorder (PTSD) among US Army combat arms soldiers using survey data collected before deployment. Methods A sample of n = 2542 US Army combat arms soldiers completed a survey shortly before deployment to Afghanistan and then again two to three and 8–9 months after redeployment. Retrospective self‐reports were obtained about CRS. Precision treatment methods were used to determine whether differential resilience to persistent PTSD in the follow‐up surveys could be developed from pre‐deployment survey data in a 60% training sample and validated in a 40% test sample. Results 40.8% of respondents experienced high CRS and 5.4% developed persistent PTSD. Significant test sample heterogeneity was found in resilience (t = 2.1, p = 0.032), with average treatment effect (ATE) of high CRS in the 20% least resilient soldiers of 17.1% (SE = 5.5%) compared to ATE = 3.8% (SE = 1.2%) in the remaining 80%. The most important predictors involved recent and lifetime pre‐deployment distress disorders. Conclusions A reliable pre‐deployment resilience score can be constructed to predict variation in the effects of high CRS on persistent PTSD among combat arms soldiers. Such a score could be used to target preventive interventions to reduce PTSD or other resilience‐related outcomes.


An Anatomy of Event Studies: Hypothetical Experiments, Exact Decomposition, and Weighting Diagnostics

October 2024

·

24 Reads

In recent decades, event studies have emerged as a central methodology in health and social research for evaluating the causal effects of staggered interventions. In this paper, we analyze event studies from the perspective of experimental design and focus on the use of information across units and time periods in the construction of effect estimators. As a particular case of this approach, we offer a novel decomposition of the classical dynamic two-way fixed effects (TWFE) regression estimator for event studies. Our decomposition is expressed in closed form and reveals in finite samples the hypothetical experiment that TWFE regression adjustments approximate. This decomposition offers insights into how standard regression estimators borrow information across different units and times, clarifying and supplementing the notion of forbidden comparison noted in the literature. We propose a robust weighting approach for estimation in event studies, which allows investigators to progressively build larger valid weighted contrasts by leveraging increasingly stronger assumptions on the treatment assignment and potential outcomes mechanisms. This weighting approach also allows for the generalization of treatment effect estimates to a target population. We provide diagnostics and visualization tools and illustrate these methods in a case study of the impact of divorce reforms on female suicide.


Figure 1: Balance of the covariate "age" by contrast value, +1 or −1, in each of six type of blocks.
Figure 2: Balance of the covariate "age × LE" or x 3 × w ′ by contrast value, +1 or −1, in each of six type of blocks.
Figure 3: Duration of unemployment, with and without an increase in benefits duration (B/b) in block type 2 at high R and block type 5 at low R, after (a) and before (b). Contrast weights, h g = 1 or h g = −1, appear at the top. The right boxplot shows 1400 = 700+700 difference-in-difference estimates, pooling blocks of types 2 and 5, with symmetric transformation of the tails to better visualize the center of the distribution.
Effect Aliasing in Observational Studies

August 2024

·

29 Reads

In experimental design, aliasing of effects occurs in fractional factorial experiments, where certain low order factorial effects are indistinguishable from certain high order interactions: low order contrasts may be orthogonal to one another, while their higher order interactions are aliased and not identified. In observational studies, aliasing occurs when certain combinations of covariates -- e.g., time period and various eligibility criteria for treatment -- perfectly predict the treatment that an individual will receive, so a covariate combination is aliased with a particular treatment. In this situation, when a contrast among several groups is used to estimate a treatment effect, collections of individuals defined by contrast weights may be balanced with respect to summaries of low-order interactions between covariates and treatments, but necessarily not balanced with respect to summaries of high-order interactions between covariates and treatments. We develop a theory of aliasing in observational studies, illustrate that theory in an observational study whose aliasing is more robust than conventional difference-in-differences, and develop a new form of matching to construct balanced confounded factorial designs from observational data.


Factors Underlying Reduced Hospitalizations for Myocardial Infarction During the COVID-19 Pandemic

July 2024

·

6 Reads

·

2 Citations

Importance The incidence of hospital encounters for acute myocardial infarction (AMI) decreased sharply early in the COVID-19 pandemic and has not returned to prepandemic levels. There has been an ongoing debate about what mechanism may underlie this decline, including patients avoiding the hospital for treatment, excess mortality from COVID-19 among patients who would otherwise have had an AMI, a reduction in the incidence or severity of AMIs due to pandemic-related changes in behavior, or a preexisting temporal trend of lower AMI incidence. Objective To describe drivers of changing incidence in AMI hospital encounters during the COVID-19 pandemic. Design, Setting, and Participants This cross-sectional study used traditional Medicare claims from all patients enrolled in traditional Medicare from January 2016 to June 2023 (total of 2.85 billion patient-months) to calculate the rate of AMI hospital encounters (emergency department visits, observation stays, or inpatient admissions) per capita at all short-term acute care and critical access hospitals in the United States overall and by patient characteristics. Observed rates were compared with expected rates that accounted for shifts in population characteristics and the prepandemic temporal trend (as estimated over 2016-2019). Data were analyzed in November 2023. Main Outcomes and Measures Hospital encounters for AMI. Results On average, the study sample included 31 623 928 patients each month from January 2016 through June 2023, for a total of 2 846 153 487 patient-months during the 90-month study period. In June 2023, there were 0.044 AMI hospital encounters per 100 patients, which was 20% lower than in June 2019 (0.055 encounters per 100 patients). Early in the pandemic, AMI rates moved inversely with COVID-19 death rates and tracked patterns seen for other painful acute conditions, such as nephrolithiasis, suggesting these changes were associated with care avoidance. Changes in patient characteristics driven by excess deaths during the pandemic explained little of the decline. Later in the pandemic, the decline may be explained by the long-standing downward trend in AMI incidence; by April 2022, the observed rate of encounters matched the expected rate that accounted for this trend. During the full pandemic period, from March 2020 to June 2023, there were an estimated 5% (95% prediction interval, 1%-9%) fewer AMI hospital encounters than expected. Conclusions and Relevance The early reduction in AMI encounters was likely driven by care avoidance, while ongoing reductions through June 2023 likely reflect long-standing temporal trends. During the pandemic, there were 5% fewer AMI encounters than expected.


The oncology care model and initiation of systemic therapy for cancer.

June 2024

·

7 Reads

Journal of Clinical Oncology

11031 Background: The CMS Oncology Care Model (OCM) was an episode payment model for patients with cancer receiving systemic therapy (chemotherapy, targeted therapy, immunotherapy, or hormonal therapy) that began in July 2016. Voluntarily participating practices received monthly care coordination payments of 160perpatientpermonthandwererequiredtodeliverenhancedservices;theywereeligibletoshareinsavingsiftheyachievedqualityandspendingtargets(basedonhistoricspendingtrendedforward;allpracticeswereinonesidedriskarrangementsbefore2019).Priorresearchidentifiedsavingsof160 per patient per month and were required to deliver enhanced services; they were eligible to share in savings if they achieved quality and spending targets (based on historic spending trended forward; all practices were in one-sided risk arrangements before 2019). Prior research identified savings of 499 per episode (excluding monthly payments) through OCM’s first 5 years, but no overall savings after including incentive payments. One concern about an episode payment model triggered by initiation of systemic therapy is that the financial incentives may prompt an increase in the total number of episodes. We assessed if OCM led to an increase in the likelihood of initiating systemic therapy. Methods: Using Medicare data, we studied care for beneficiaries enrolled in Parts A, B, and D of fee-for-service Medicare with index cancer diagnoses in 2010 through 2019. We assessed initiation of systemic therapy in the one year after the index diagnosis date. We studied two populations: (1) all patients with an index cancer diagnosis and no cancer diagnosis in the preceding two years, suggesting newly diagnosed or newly progressive cancers (incident cohort) and (2) patients with poor prognosis cancers. We used a difference-in-differences (DiD) analysis to assess systemic therapy initiation among patients with index cancer diagnoses who had office visits to medical oncology practices that were participating in OCM, compared with matched comparison practices (on number of patients attributed, number of physicians, number of medical oncologists, and academic affiliation, favoring matches within Hospital Referral Region), before and after OCM’s start in July 2016. Analyses adjusted for patient demographic and clinical variables. Results: Among 742,699 beneficiaries in the incident cancer cohort, 61.9% initiated systemic therapy within 1 year of their index diagnosis. Among 777,951 beneficiaries in the poor prognosis cohort, 58.2% initiated systemic therapy within 1 year of their index diagnosis date. OCM was not associated with the likelihood of initiating systemic therapy in the incident cohort (DID:-0.7 percentage point change, 95% CI:-1.9,0.4, P=0.19) or the poor prognosis cohort (DID:-1.0 percentage point change, 95% CI:-2.1,0.1, P=0.07). Conclusions: Despite financial incentives of episode payment models that may favor greater use of systemic therapy for patients with cancer, OCM did not increase the likelihood of initiating systemic therapy episodes among patients with newly diagnosed/newly progressive cancers or poor prognosis cancers who visited a medical oncology practice.



Preferences for Postacute Care at Home vs Facilities

April 2024

·

5 Reads

JAMA Health Forum

Importance Two in 5 US hospital stays result in rehabilitative postacute care, typically through skilled nursing facilities (SNFs) or home health agencies (HHAs). However, a lack of clear guidelines and understanding of patient and caregiver preferences make it challenging to promote high-value patient-centered care. Objective To assess preferences and willingness to pay for facility-based vs home-based postacute care among patients and caregivers, considering demographic variations. Design, Setting, and Participants In September 2022, a nationally representative survey was conducted with participants 45 years or older. Using a discrete choice experiment, participants acting as patients or caregivers chose between facility-based and home-based postacute care that best met their preferences, needs, and family conditions. Survey weights were applied to generate nationally representative estimates. Main Outcomes and Measures Preferences and willingness to pay for various attributes of postacute care settings were assessed, examining variation based on demographic factors, socioeconomic status, job security, and previous care experiences. Results A total of 2077 adults were invited to participate in the survey; 1555 (74.9%) completed the survey. In the weighted sample, 52.9% of participants were women, 6.5% were Asian or Pacific Islander, 1.7% were American Indian or Alaska Native, 11.2% were Black or African American, 78.4% were White; the mean (SD) age was 62.6 (9.6) years; and there was a survey completion rate of 74.9%. Patients and caregivers showed a substantial willingness to pay for home-based and high-quality care. Patients and caregivers were willing to pay an additional 58.08perday(9558.08 per day (95% CI, 45.32-70.83) and 45.54 per day (95% CI, 31.09-59.99) for HHA care compared with a shared SNF room, respectively. However, increased demands on caregiver time within an HHA scenario and socioeconomic challenges, such as insecure employment, shifted caregivers’ preferences toward facility-based care. There was a strong aversion to below average quality. To avoid below average SNF care, patients and caregivers were willing to pay 75.21perday(9575.21 per day (95% CI, 61.68-88.75) and 79.10 per day (95% CI, 63.29-94.91) compared with average-quality care, respectively. Additionally, prior awareness and experience with postacute care was associated with willingness to pay for home-based care. No differences in preferences among patients and caregivers based on race, educational background, urban or rural residence, general health status, or housing type were observed. Conclusions and Relevance The findings of this survey study underscore a prevailing preference for home-based postacute care, aligning with current policy trends. However, attention is warranted for disadvantaged groups who are potentially overlooked during the shift toward home-based care, particularly those facing caregiver constraints and socioeconomic hardships. Ensuring equitable support and improved quality measure tools are crucial for promoting patient-centric postacute care, with emphasis on addressing the needs of marginalized groups.


Causation, Comparison, and Regression

January 2024

·

27 Reads

·

2 Citations

Comparison and contrast are the basic means to unveil causation and learn which treatments work. To build good comparison groups, randomized experimentation is key, yet often infeasible. In such non-experimental settings, we illustrate and discuss diagnostics to assess how well the common linear regression approach to causal inference approximates desirable features of randomized experiments, such as covariate balance, study representativeness, interpolated estimation, and unweighted analyses. We also discuss alternative regression modeling, weighting, and matching approaches and argue they should be given strong consideration in empirical work.


Lmw: Linear Model Weights for Causal Inference

January 2024

·

6 Reads

·

1 Citation

Observational Studies

The linear regression model is widely used to adjust for covariates and estimate the average effects of treatments. Behind every causal inference endeavor there is a hypothetical randomized experiment. However, in routine regression analyses in observational studies, it is unclear how well the adjustments made by regression approximate key features of randomized experiments, such as covariate balance, study representativeness, sample boundedness, and unweighted sampling. In this paper, we provide software to empirically address this question. We introduce the lmw package for R to compute the implied linear model weights and perform related diagnostics. The weights are obtained as part of the design stage of the study; that is, without using outcome information. The implementation is general and applicable, for instance, in settings with instrumental variables and multi-valued treatments; in essence, in any situation where the linear model is the vehicle for adjustment and estimation of average treatment effects with discrete-valued interventions.


Citations (64)


... Furthermore, we connect the 2sls estimator with the reweighting perspective in Abadie (2003), and establish the necessity of condition (i) in the absence of additional assumptions on potential outcomes. The results extend the literature on the connection between regression and reweighting for estimating the average treatment effect and LATE (Kline, 2011;Chattopadhyay and Zubizarreta, 2023;Chattopadhyay et al., 2024;S loczyński et al., 2025). Third, to remedy the possible biases when neither condition (i) nor condition (ii) holds, we propose stratifying based on the estimated propensity score of the IV and using the stratum dummies as the new covariates to fit the interacted 2sls. ...

Reference:

Interacted two-stage least squares with treatment effect heterogeneity
Lmw: Linear Model Weights for Causal Inference
  • Citing Article
  • January 2024

Observational Studies

... The overall concept of 3PM medicine is gaining upper hands in the medical revolution since scientists are looking it through the multiomics technology and advanced framework of research [33]. One case of diseases complications where real-time monitoring is applicable is the introducing of the 3PM on the management of stroke [34]. In a way of targeted preventive intervention, a composite score while evaluating the differential resilience based on the effect on combat-related stress (CRS) is great tool for neuroscience. ...

A prediction model for differential resilience to the effects of combat‐related stressors in US army soldiers

... It is worth noting that the significant decline in CAD and AMI mortality over the past 2 decades predominantly occurred in the first decade, with a notable deceleration after 2011. This is likely due to AMI. [26][27][28] This rise in cardiovascular deaths, including those from ischemic heart disease, was particularly pronounced in regions heavily affected by COVID-19. [26][27][28][29] Other effects of the COVID-19 ...

Factors Underlying Reduced Hospitalizations for Myocardial Infarction During the COVID-19 Pandemic
  • Citing Article
  • July 2024

... Indirect standardization, on the other hand, compares the outcomes of patients treated by given provider to their counterfactual outcomes had they been randomly reassigned to another provider according to a specified probability distribution. There is some flexibility in defining the counterfactual reassignment mechanism: patients could be assigned to a provider uniformly at random, to a provider with "average" performance (He et al., 2013), to a hypothetical provider formed by pooling patients across all providers or consolidating providers within a certain geographical region (Varewyck et al., 2014;Han et al., 2024), or to a provider that tends to treat patients with similar covariates. By carefully choosing the randomization process, we can avoid the positivity problem inherent in direct standardization (Daignault and Saarela, 2017). ...

Privacy-preserving, communication-efficient, and target-flexible hospital quality measurement
  • Citing Article
  • June 2024

The Annals of Applied Statistics

... The causal inference literature has increasingly acknowledged the importance of not solely relying on ordinary least squares (OLS) regression and comparing OLS estimates with those derived from other methods [42][43][44] . Following relevant recommendations, we compared regression and PSM estimates across various models ( Fig. 1 and Supplementary Fig. 10). ...

Causation, Comparison, and Regression

... Although φ is typically infinite-dimensional, the Riesz Representer Theorem shows that the least square regression and, equivalently, the balancing optimization problem have closed-form solutions. The least squares regression approach is kernel ridge regression and the weighting estimator is kernel balancing weights (see Hazlett, 2020;Kim et al., 2022a). Hirshberg et al. (2019) leverage this equivalence to analyze the asymptotic bias of kernel balancing weights. ...

Scalable kernel balancing weights in a nationwide observational study of hospital profit status and heart attack outcomes
  • Citing Article
  • December 2023

Biostatistics

... Several earlier studies report significant benefits associated with RPM. For example, RPM usage has been linked with improved blood pressure outcomes among patients with hypertension, [4][5][6][7] while similar improvements were seen in RPM interventions to promote blood sugar control among patients with diabetes. 8 RPM has also been linked to reduced mortality and readmissions for patients with heart failure and chronic obstructive pulmonary disease. ...

Effects of Remote Patient Monitoring Use on Care Outcomes Among Medicare Patients With Hypertension : An Observational Study
  • Citing Article
  • November 2023

Annals of Internal Medicine

... A growing body of evidence makes clear that exposure to firearm violence is associated with poorer mental, physical, and behavioral health among individuals Smith et al., 2020;Song et al., 2022Song et al., , 2023Turner et al., 2019;Vasan et al., 2021) and throughout entire communities (Semenza & Stansfield 2021a, b;Semenza et al., 2023b). In two recent longitudinal studies using case-control matches of commercial health insurance claims, researchers found that children and adolescent survivors of firearm violence experienced substantial increases in pain, psychiatric, and substance use disorders in the year following firearm injury relative to controls (Song et al., 2023). ...

Firearm Injuries In Children And Adolescents: Health And Economic Consequences Among Survivors And Family Members
  • Citing Article
  • November 2023

Health Affairs

... Individuals whose medication adherence decreased were less likely to transition from in-person care to telehealth and less likely to receive case management services and were more likely to visit the ED and to be hospitalized. This finding stands in contrast to a prior study, which found that reduced outpatient and inpatient visits were not associated with reduced adherence (Wilcock et al., 2023), although the sample for that study included only Medicare beneficiaries. The fact that 70% remained adherent to their medication regimen highlights both their personal strength and the adaptability of the Philadelphia publicly funded treatment system, which implemented creative and flexible approaches to care, including a capitated payment that maintained the service system and offered agencies flexibility in providing care. ...

Use of Telemedicine and Quality of Care Among Medicare Enrollees With Serious Mental Illness

JAMA Health Forum

... In addition to identifying the transportable population as a subset of the target population, entropy weighting and its modifications have been proposed as alternatives that identify the transportable population using moment constraints to align the mean or other aspects of the covariate distribution [38,39]. However, entropy weighting relies on untestable parametric assumptions about the outcome given the covariates. ...

One-Step Weighting to Generalize and Transport Treatment Effect Estimates to a Target Population
  • Citing Article
  • October 2023