Guanghao Zhang

The performance of prediction algorithms is typically measured using four metrics: sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). These metrics are usually calculated on samples drawn from patient populations. However, the performance metrics computed over a deliberately biased sample would not directly extend to its source population. Further, it is often necessary to infer the metric values for a population different from where the sample was drawn. In this paper, we illustrate methods to solve both challenges. Specifically, given the underlying patient distribution, we show corrections to the formula for these metrics based on two common inverse probability weighting methods: standard cell weighting and logistic regression weighting. We conduct simulation experiments to identify patients living with dementia and compare these methods in performance corrections with different sample sizes for different prevalence settings. We empirically show that weighting methods can correct the estimated values for algorithms’ performance. Standard cell weighting is preferred over logistic regression weighting when the sample size is small and only the strata information is available in the populations of interest.

ObjectivesInvestigate the association between advance care planning and decision regret among stroke surrogate decision makers.MethodsPeople hospitalized with ischemic or hemorrhagic stroke and the...

Stroke has a large aggregate financial effect, but the prevalence and predictors of cost concerns among family surrogate decision-makers of patients with stroke are unknown. The aim of this study was to explore the prevalence, predictors, and nature of cost concerns among family surrogate decision-makers of patients with stroke. Family surrogate decision-makers of patients with stroke were recruited from a largely biethnic population-based study in Nueces County, TX. The primary outcome was the presence of cost concerns, defined as being "very worried" or "moderately worried" about being able to pay stroke-related medical costs, assessed through surveys conducted at baseline (shortly after stroke) and 3-, 6-, and 12-month poststroke. Potential predictors included clinical and sociodemographic factors for patients and family members, obtained from surveys and medical charts. Exploratory multivariable logistic regression models identified factors associated with cost concerns, adjusting for covariates. We also explored the nature of cost concerns in semistructured interviews with a subset of participants. Of 465 eligible stroke cases, 318 family members of 256 patients with stroke participated. Patients were 52% female, 58% Mexican American (MA), 38% non-Hispanic White (NHW), and 8% other race/ethnicity, mean age 75 (SD 14), and median NIH Stroke Scale score of 12.5 (IQR 5-22). Family members were 76% female, 63% MA, 32% NHW, and 5% other race/ethnicity, mean age 56 (SD 13). Of 256 patients, 118 (46%) had a family member with cost concerns. After adjustment for all covariates, MA ethnicity and lack of insurance were associated with greater cost concerns. Cost concerns decreased over time and were less common among family members who were neither a spouse nor a child of the patient. In semistructured interviews, cost concerns related to postacute care were most frequently mentioned, but concerns regarding hospital costs and nonmedical costs were also common. Nearly half of family surrogate decision-makers of patients with stroke had at least moderate concerns about being able to afford stroke-related medical costs. Lack of adequate insurance and membership in a historically marginalized ethnic group were the strongest predictors of cost concerns.

Electronic health records (EHRs) are increasingly recognized as a cost-effective resource for patient recruitment in clinical research. However, how to optimally select a cohort from millions of individuals to answer a scientific question of interest remains unclear. Consider a study to estimate the mean or mean difference of an expensive outcome. Inexpensive auxiliary covariates predictive of the outcome may often be available in patients’ health records, presenting an opportunity to recruit patients selectively, which may improve efficiency in downstream analyses. In this paper we propose a two-phase sampling design that leverages available information on auxiliary covariates in EHR data. A key challenge in using EHR data for multiphase sampling is the potential selection bias, because EHR data are not necessarily representative of the target population. Extending existing literature on two-phase sampling design, we derive an optimal two-phase sampling method that improves efficiency over random sampling while accounting for the potential selection bias in EHR data. We demonstrate the efficiency gain from our sampling design via simulation studies and an application evaluating the prevalence of hypertension among U.S. adults leveraging data from the Michigan Genomics Initiative, a longitudinal biorepository in Michigan Medicine.

BACKGROUND:Patients with severe stroke often rely on surrogate decision-makers for life-sustaining treatment decisions. We investigated ethnic differences between Mexican Americans (MAs) and non-Hispanic White (NHW) individuals in surrogate reports of physician quality of communication and shared decision-making from the OASIS study (Outcomes Among Surrogate Decision Makers in Stroke) project.METHODS:Patients had ischemic stroke or intracerebral hemorrhage in Nueces County, TX. Surrogates self-identified as being involved in decisions about do-not-resuscitate orders, brain surgery, ventilator, feeding tube, or hospice/comfort care. Surrogate reports of physician quality of communication (scale score, range from 0 to 10) and shared decision-making (CollaboRATE scale score, binary score 1 versus 0) were compared by ethnicity with linear or logistic regression using generalized estimating equations, adjusted for prespecified demographics, clinical factors, and confounders.RESULTS:Between April 2016 and September 2020, 320 surrogates for 257 patients with stroke enrolled (MA, 158; NHW, 85; and other, 14). Overall quality of communication score was better among surrogates of MA patients than NHW individuals after adjustment for demographics, stroke severity, and patient survival though the ethnic difference was attenuated (beta, 0.47 [95% CI, -0.17 to 1.12]; P=0.15) after adjustment for trust in the medical profession and frequency of personal prayer. High CollaboRATE scale scores were more common among surrogates of MA patients than NHW individuals (unadjusted odds ratio, 1.75 [95% CI, 1.04-2.95]). This association persisted after adjustment for demographic and clinical factors though there was an interaction between patient age and ethnicity (P=0.04), suggesting that this difference was primarily in older patients.CONCLUSIONS:Surrogate decision-makers of MA patients generally reported better outcomes on validated measures of quality of communication and shared decision-making than NHW individuals. Further study of outcomes among diverse populations of stroke surrogate decision-makers may help to identify sources of strength and resiliency that may be broadly applicable.

During acute hospitalizations, physicians often focus on the stroke patient and not family who may be traumatized by this sudden change to their loved one. We investigated long-term psychological distress among family surrogate decision makers for Mexican American (MA) and non-Hispanic White (NHW) severe stroke patients. Previous work in other diseases suggested worse psychological outcomes in MA than NHW caregivers. This was a population-based, prospective cohort study in Nueces County, TX. Stroke patient participants and their surrogate decision makers were enrolled soon after any stroke between April, 2016, and October, 2020, if surrogates had made decisions about life-sustaining treatments. Surrogates completed validated measures of posttraumatic stress, National Stressful Events Survey for Posttraumatic Stress Disorder Short Scale; anxiety, Generalized Anxiety Disorder-7; and depression, Patient Health Questionnaire-8 at discharge, 3, 6, and 12 months. Ethnic differences were assessed with multilevel linear mixed models, sequentially adjusted for prespecified patient and surrogate demographic, socioeconomic, and clinical covariates. There were 301 family surrogates for 241 severe stroke patients. The mean follow-up was 315 days. High scores on measures of psychological distress ranged between 17% and 28% of surrogates. One or more high levels of the psychological outcomes were found in 17%-43% of surrogates; 2 or more were found in 12%-27%; and all 3 were found in 5%-16% of surrogates. All psychological outcomes were worse among MAs on unadjusted analyses. In fully adjusted models, posttraumatic stress remained worse among MAs (0.36, 95% CI 0.17-0.56); ethnic differences were attenuated and no longer significant in the final model for anxiety (0.59, 95% CI -0.55 to 1.74) and depression (0.97, 95% CI -0.25 to 2.19). The trajectory for depression did differ by ethnicity (interaction p = 0.03), with depression score improving more rapidly over time among NHWs than MAs. Advance care plans did not seem to confound any ethnic differences. Psychological distress is common among family surrogate decision makers in the year after stroke and may be worse among MAs. Efforts are needed to support family members of all ethnic groups after severe stroke.

Electronic health records (EHRs) are increasingly recognized as a cost-effective resource for patient recruitment in clinical research. However, how to optimally select a cohort from millions of individuals to answer a scientific question of interest remains unclear. Consider a study to estimate the mean or mean difference of an expensive outcome. Inexpensive auxiliary covariates predictive of the outcome may often be available in patients' health records, presenting an opportunity to recruit patients selectively which may improve efficiency in downstream analyses. In this paper, we propose a two-phase sampling design that leverages available information on auxiliary covariates in EHR data. A key challenge in using EHR data for multi-phase sampling is the potential selection bias, because EHR data are not necessarily representative of the target population. Extending existing literature on two-phase sampling design, we derive an optimal two-phase sampling method that improves efficiency over random sampling while accounting for the potential selection bias in EHR data. We demonstrate the efficiency gain from our sampling design via simulation studies and an application to evaluating the prevalence of hypertension among US adults leveraging data from the Michigan Genomics Initiative, a longitudinal biorepository in Michigan Medicine.

BACKGROUND: The aim of this study was to characterize change in sleep-disordered breathing severity in the year following stroke, overall, and by ethnicity, within the population-based Brain Attack Surveillance in Corpus Christi Project. METHODS: First-ever ischemic strokes (n=414) were ascertained by active and passive surveillance and validated by stroke-trained physicians. Patients with stroke were invited to participate in portable sleep apnea testing (ApneaLink Plus) at baseline and 3, 6, and 12 months poststroke. Sleep-disordered breathing severity was assessed by the respiratory event index (apneas and hypopneas/hour of recording). The component obstructive apnea index and central apnea index were also assessed. Time and ethnicity effects on outcomes, as well as ethnic differences in time effects, were analyzed using generalized estimating equations with multivariable adjustment for confounding factors. RESULTS: Mean age (n=414) was 63.9 years (SD=10.9); 68.4% were Mexican American. Baseline mean respiratory event index, obstructive apnea index, and central apnea index were 21.3 (SD=16.6), 8.6 (SD=11.5), and 1.5 (SD=3.2), respectively. There was no time effect on respiratory event index ( P =0.35) but obstructive apnea index increased over time ( P P =0.01) such that central apnea index increased in non-Hispanic White but did not change in Mexican American persons. CONCLUSIONS: Sleep-disordered breathing severity was significant and stable for most individuals in the year after stroke. These results do not confirm previous reports of diminishing sleep-disordered breathing severity over time after stroke and would support early assessment and treatment where indicated.