Search

The Countrywide Mortality Surveillance for Action platform is collecting verbal autopsy (VA) records from a nationally representative sample in Mozambique. These records are used to estimate the national and subnational cause-specific mortality fractions (CSMFs) for children (1–59 months) and neonates (1–28 days). Cross-tabulation of VA-based cause-of-death (COD) determination against that from the minimally invasive tissue sampling (MITS) from the Child Health and Mortality Prevention project revealed important misclassification errors for all the VA algorithms, which if not accounted for will lead to bias in the estimates of CSMF from VA. A recently proposed Bayesian VA-calibration method is used that accounts for this misclassification bias and produces calibrated estimates of CSMF. Both the VA-COD and the MITS-COD can be multi-cause (i.e., suggest more than one probable COD for some of the records). To fully use this probabilistic COD data, we use the multi-cause VA calibration. Two different computer-coded VA algorithms are considered—InSilicoVA and EAVA—and the final CSMF estimates are obtained using an ensemble calibration that uses data from both the algorithms. The calibrated estimates consistently offer a better fit to the data and reveal important changes in the CSMF for both children and neonates in Mozambique after accounting for VA misclassification bias.

Sub-Saharan Africa lacks timely, reliable, and accurate national data on mortality and causes of death (CODs). In 2018 Mozambique launched a sample registration system (Countrywide Mortality Surveillance for Action [COMSA]-Mozambique), which collects continuous birth, death, and COD data from 700 randomly selected clusters, a nationally representative population of 828,663 persons. Verbal and social autopsy interviews are conducted for COD determination. We analyzed data collected in 2019–2020 to report mortality rates and cause-specific fractions. Cause-specific results were generated using computer-coded verbal autopsy (CCVA) algorithms for deaths among those age 5 years and older. For under-five deaths, the accuracy of CCVA results was increased through calibration with data from minimally invasive tissue sampling. Neonatal and under-five mortality rates were, respectively, 23 (95% CI: 18–28) and 80 (95% CI: 69–91) deaths per 1,000 live births. Mortality rates per 1,000 were 18 (95% CI: 14–21) among age 5–14 years, 26 (95% CI: 20–31) among age 15–24 years, 258 (95% CI: 230–287) among age 25–59 years, and 531 (95% CI: 490–572) among age 60+ years. Urban areas had lower mortality rates than rural areas among children under 15 but not among adults. Deaths due to infections were substantial across all ages. Other predominant causes by age group were prematurity and intrapartum-related events among neonates; diarrhea, malaria, and lower respiratory infections among children 1–59 months; injury, malaria, and diarrhea among children 5–14 years; HIV, injury, and cancer among those age 15–59 years; and cancer and cardiovascular disease at age 60+ years. The COMSA-Mozambique platform offers a rich and unique system for mortality and COD determination and monitoring and an opportunity to build a comprehensive surveillance system.

Associations of mean glucose and time in range (70-180 mg/dL) from continuous glucose monitoring (CGM) with HbA1c in adults with type 2 diabetes are not well characterized. We conducted a secondary analysis of 186 participants from the Hyperglycemic Profiles in Obstructive Sleep Apnea (HYPNOS) trial. Participants simultaneously wore Dexcom G4 and Abbott Libre Pro CGM sensors up to 4 weeks. Mean HbA1c was 7.7% (SD, 1.3). There were strong negative Pearson's correlations of HbA1c with CGM time in range (-0.79, Abbott; -0.81, Dexcom) and strong positive correlations with CGM mean glucose (Dexcom, 0.84; Abbott, 0.82). However, there were large differences in CGM mean glucose (±20 mg/dL) and time in range (±14%) at any given HbA1c value. Mean glucose and HbA1c are strongly correlated in type 2 diabetes patients not taking insulin but discordance is evident at the individual level. Clinicians should expect discordance and use HbA1c and CGM in a complementary manner. ClinicalTrials.gov Identifier: NCT02454153.

In the management of symptomatic bone metastases, current practice guidelines do not provide clear methodology for selecting palliative radiotherapy (RT) regimens based on specific patient and disease features. Decision support aids may offer an effective means for translating the complex data needed to render individualised treatment decisions, yet no such tools are available for use in this setting. Thus, we describe the development of the Bone Metastases Ensemble Trees for Survival-Decision Support Platform (BMETS-DSP), which aims to optimise selection of evidence-based, individualised palliative RT regimens.The Ottawa Decision Support Framework was used as the theoretical basis for development of BMETS-DSP. First, we utilised stakeholder input and review of the literature to assess determinants underlying the provider decision. Based on this assessment and iterative stakeholder feedback, we developed the web-based, provider-facing BMETS-DSP. Consistent with the underlying theoretical framework, our design also included assessment of decision quality using the International Patient Decision Aids Standards (IPDAS) certification checklist.Stakeholder input and review of 54 evidence-based publications identified the following determinants of the provider decision: estimated prognosis, characteristics of the target symptomatic lesion and the primary cancer type, consideration of alternative interventions, access to patient-specific recommendations, and patient preferences. Based on these determinants, we developed the BMETS-DSP that (1) collects patient-specific data, (2) displays an individualised predicted survival curve, and (3) provides case-specific, evidence-based recommendations regarding RT, open surgery, systemic therapy, and hospice referral to aid in the decision-making process. The finalised tool met IPDAS quality requirements. Preliminary results of a pilot assessment suggest impact of clinical outcomes.We describe the successful development of a provider-facing decision support platform to aid in the provision of palliative RT in better alignment with patient and disease features. Impact of the BMETS-DSP on decision outcomes will be further assessed in a randomised, controlled study.

Introduction: Continuous glucose monitoring (CGM) devices record blood glucose every few minutes over a maximum 2-week wear period. Despite the richness of these data, existing first and second moment statistics (such as mean glucose) fail to capture the “shape” of variation in the CGM time-series. Objective: We developed novel, reproducible CGM shape metrics using functionals of the periodogram, a decomposition of CGM signal ( Figure 1A ) variance, to comprehensively characterize glucose patterns. Methods: We analyzed CGM data from N = 160 HYPNOS clinical trial participants with type 2 diabetes who wore CGM sensors twice, 3 months apart. We first calculated log-transformed periodograms for each person-period. To these decompositions, we fit piece-wise linear models over 3 frequency ranges: less than 1/24 hrs -1 (long term patterns), 1/24 to 2/5 hrs -1 (daily diet patterns), and greater than 2/5 hrs -1 (immediate fluctuations due to food). Slopes of fit segments, value at first frequency, and values at segment midpoints were extracted as our initial metrics ( Figure 1B ). We estimated and compared the within-person test-retest correlation for our shape measures against existing CGM and glucose metrics: mean glucose, time-in-range, and HhbA1c. Results: The highest raw correlation among the new shape metrics (r = 0.737) was comparable to that of the existing metrics (r = 0.798) ( Figure 1C ). Even after adjusting the shape metrics for existing metrics, there was a maximum test-retest correlation of r = 0.778. The shape metrics also carried information distinct from existing metrics. According to mixed effect models fit to data from both wear periods, the shape metrics explain less than 16% of variance in any existing metric when controlled for the other existing metrics. Conclusion: These new shape metrics carry reproducible information distinct from that of standard CGM metrics and HbA1c. The next step is to evaluate whether these novel shape metrics, which leverage the granularity of CGM data, can be linked to clinical endpoints.

Background Despite growing interest in monitoring improvements in quality of care, data on service quality in low-income and middle-income countries (LMICs) is limited. While health systems researchers have hypothesized the relationship between facility readiness and provision of care, there have been few attempts to quantify this relationship in LMICs. This study assesses the association between facility readiness and provision of care for antenatal care at the client level and facility level. Methods To assess the association between provision of care and various facility readiness indices for antenatal care, we used multilevel, multivariable random-effects linear regression models. We tested an inflection point on readiness scores by fitting linear spline models. To compare the coefficients between models, we used a bootstrapping approach and calculated the mean difference between all pairwise comparisons. Analyses were conducted at client and facility levels. Results Our results showed a small, but significant association between facility readiness and provision of care across countries and most index constructions. The association was most evident in the client-level analyses that had a larger sample size and were adjusted for factors at the facility, health worker, and individual levels. In addition, spline models at a facility readiness score of 50 better fit the data, indicating a plausible threshold effect. Conclusions The results of this study suggest that facility readiness is not a proxy for provision of care, but that there is an important association between facility readiness and provision of care. Data on facility readiness is necessary for understanding the foundations of health systems particularly in countries with the lowest levels of service quality. However, a comprehensive view of quality of care should include both facility readiness and provision of care measures.

Objectives Ectopic calcification (calcinosis) is a common complication of SSc, but a subset of SSc patients has a heavy burden of calcinosis. We examined whether there are unique risk factors for a heavy burden of calcinosis, as compared with a light burden or no calcinosis. Methods We reviewed the medical records of all patients in the Johns Hopkins Scleroderma Center Research Registry with calcinosis to quantify calcinosis burden using pre-specified definitions. We performed latent class analysis to identify SSc phenotypic classes. We used multinomial logistic regression to determine whether latent phenotypic classes and autoantibodies were independent risk factors for calcinosis burden. Results Of all patients, 29.4% (997/3388) had calcinosis, and 13.5% (130/963) of those with calcinosis had a heavy burden. The latent phenotypic class with predominantly diffuse skin disease and higher disease severity (characterized by pulmonary hypertension, interstitial lung disease, cardiomyopathy, severe RP, gastrointestinal involvement, renal crisis, myopathy and/or tendon friction rubs) was associated with an increased risk of both a heavy burden [odds ratio (OR) 6.92, 95% CI 3.66, 13.08; P Conclusions Calcinosis burden is associated with cumulative SSc-related tissue damage. Independent of disease severity, autoantibodies to PM/Scl are also associated with a heavy burden of calcinosis.

PURPOSE The Bone Metastases Ensemble Trees for Survival Decision Support Platform (BMETS-DSP) provides patient-specific survival predictions and evidence-based recommendations to guide multidisciplinary management for symptomatic bone metastases. We assessed the clinical utility of the BMETS-DSP through a pilot prepost design in a simulated clinical environment. METHODS Ten Radiation Oncology physicians reviewed 55 patient cases at two time points: without and then with the use of BMETS-DSP. Assessment included 12-month survival estimate, confidence in and likelihood of sharing estimates with patients, and recommendations for open surgery, systemic therapy, hospice referral, and radiotherapy (RT) regimen. Paired statistics compared pre- versus post-DSP outcomes. Reported statistical significance is P < .05. RESULTS Pre- versus post-DSP, overestimation of true minus estimated survival time was significantly reduced (mean difference –2.1 [standard deviation 4.1] v –1 month [standard deviation 3.5]). Prediction accuracy was significantly improved at cut points of < 3 (72 v 79%), ≤ 6 (64 v 71%), and ≥ 12 months (70 v 81%). Median ratings of confidence in and likelihood of sharing prognosis significantly increased. Significantly greater concordance was seen in matching use of 1-fraction RT with the true survival < 3 months (70 v 76%) and < 10-fraction RT with the true survival < 12 months (55 v 62%) and appropriate use of open surgery (47% v 53%), without significant changes in selection of hospice referral or systemic therapy. CONCLUSION This pilot study demonstrates that BMETS-DSP significantly improved physician survival estimation accuracy, prognostic confidence, likelihood of sharing prognosis, and use of prognosis-appropriate RT regimens in the care of symptomatic bone metastases, supporting future multi-institutional validation of the platform.

Background Estimates of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) seroprevalence in young children and risk factors for seropositivity are scarce. Using data from a prospective cohort study of households during the pre-coronavirus disease 2019 (COVID-19) vaccine period, we estimated SARS-CoV-2 seroprevalence by age and evaluated risk factors for SARS-CoV-2 seropositivity. Methods The SARS-CoV-2 Epidemiology and Response in Children (SEARCh) study enrolled 175 Maryland households (690 participants) with ≥1 child aged 0–4 years during November 2020–March 2021; individuals vaccinated against COVID-19 were ineligible. At enrollment, participants completed questionnaires about sociodemographic and health status and work, school, and daycare attendance. Participants were tested for SARS-CoV-2 antibodies in sera. Logistic regression models with generalized estimating equations (GEE) to account for correlation within households assessed predictors of individual- and household-level SARS-CoV-2 seropositivity. Results Of 681 (98.7%) participants with enrollment serology results, 55 (8.1%; 95% confidence interval [CI], 6.3%–10.4%) participants from 21 (12.0%) households were seropositive for SARS-CoV-2. Among seropositive participants, fewer children than adults reported being tested for SARS-CoV-2 infection before enrollment (odds ratio [OR] = 0.23; 95% CI, .06–.73). Seropositivity was similar by age (GEE OR vs 0–4 years: 1.19 for 5–17 years, 1.36 for adults; P = .16) and was significantly higher among adults working outside the home (GEE adjusted OR = 2.2; 95% CI, 1.1–4.4) but not among children attending daycare or school. Conclusions Before study enrollment, children and adults in this cohort had similar rates of SARS-CoV-2 infection as measured by serology. An adult household member working outside the home increased a household's odds of SARS-CoV-2 infection, whereas a child attending daycare or school in person did not.

Background The within-person and between-sensor variability of metrics from different interstitial continuous glucose monitoring (CGM) sensors in adults with type 2 diabetes not taking insulin is unclear. Methods Secondary analysis of data from 172 participants from the Hyperglycemic Profiles in Obstructive Sleep Apnea randomized clinical trial. Participants simultaneously wore Dexcom G4 and Abbott Libre Pro CGM sensors for up to 2 weeks at baseline and again at the 3-month follow-up visit. Results At baseline (up to 2 weeks of CGM), mean glucose for both the Abbott and Dexcom sensors was approximately 150 mg/dL (8.3 mmol/L) and time in range (70180 mg/dL [3.910.0 mmol/L]) was just below 80. When comparing the same sensor at 2 different time points (two 2-week periods, 3 months apart), the within-person coefficient of variation (CVw) in mean glucose was 17.4 (Abbott) and 14.2 (Dexcom). CVw for percent time in range: 20.1 (Abbott) and 18.6 (Dexcom). At baseline, the Pearson correlation of mean glucose from the 2 sensors worn simultaneously was r 0.86, root mean squared error (RMSE), 13 mg/dL (0.7 mmol/L); for time in range, r 0.88, RMSE, 8 percentage points. Conclusions Substantial variation was observed within sensors over time and across 2 different sensors worn simultaneously on the same individuals. Clinicians should be aware of this variability when using CGM technology to make clinical decisions. ClinicalTrials.gov Identifier: NCT02454153.

BackgroundIn South Asia, a third of babies are born small-for-gestational age (SGA) accounting for a quarter of all neonatal deaths. The risk factors are well described in the literature, but many studies are in high-and-middle income countries or measure SGA on facility births only. There are fewer studies that describe the prevalence of risk factors for large-for-gestational age (LGA) in low-income countries. We aim to describe the factors associated with SGA and LGA in a population-based cohort of pregnant women in rural Nepal. MethodsThis is a secondary data analysis of a community-based trial where women were assessed for pregnancy approximately every five weeks. Demographic, socio-economic status (SES), medical/obstetric history, and timing of last menstruation were collected on pregnant women at enrollment. Vital signs, illness symptoms, and antenatal care (ANC) attendance were collected throughout the pregnancy and neonatal weight was measured for live births. We conducted multivariate analysis using multinomial logistic regression. Outcomes were SGA, LGA compared to appropriate-for-gestational age (AGA) and were multiply imputed using birthweight recalibrated to time 0, at delivery.Results In the adjusted analysis, SGA was associated with lower SES, zero parity, interpregnancy intervals less than 18 months, poor appetite/vomiting in the second trimester, and swelling of hands/face in the third trimester. Parity greater than five, male fetal sex, and increased weight gain were protective for SGA. Many of these factors were inversely associated with LGA. Additionally, four or more ANC visits and respiratory symptoms in the third trimester were negatively associated with LGA, and maternal age Conclusions Our findings are in line with known risk factors for SGA. Because the prevalence and mortality risk of LGA babies is low in this population, it is likely LGA status does not indicate underlaying illness. High quality antenatal care, monitoring for appropriate gestational weight gain and increased monitoring of women with high-risk pregnancies may reduce prevalence and improve outcomes of SGA babies. Trial RegistrationThe study used in this secondary data analysis was registered at Clinicaltrials.gov NCT01177111.

As clinicians are faced with a deluge of new information, data science can play a key role in highlighting key features towards developing new clinical hypotheses. Indeed, insights derived from machine learning can serve as a clinical support tool by connecting care providers with results from big data analysis to identify latent patterns that may not be easily detected by even skilled human observers. In this work, we show an example of collaboration between clinicians and data scientists during the COVID-19 pandemic, identifying subgroups of COVID-19 patients with unanticipated outcomes or who are high-risk for severe disease or death. We apply a random forest classifier model to predict adverse patient outcomes early in the disease course, and we connect our classification results to unsupervised clustering of patient features that may underpin patient risk. The paradigm for using data science for hypothesis generation and clinical decision support, as well as our triage classification approach and unsupervised clustering methods to determine patient cohorts, are applicable to driving rapid hypothesis generation and iteration in a variety of clinical challenges, including future public health crises.

ObjectivePreterm birth can have short-term and long-term complications for a child. Socioeconomic factors and pregnancy-related morbidities may be important to predict and prevent preterm births in low-resource settings. The objective of our study was to find prevalence and predictors of spontaneous preterm birth in rural Nepal.DesignThis is a secondary observational analysis of trial data (registration numberNCT01177111).SettingRural Sarlahi district, Nepal.Participants40 119 pregnant women enrolled from 9 September 2010 to 16 January 2017.Outcome measuresThe outcome variable is spontaneous preterm birth. Generalized Estimating Equations Poisson regression with robust variance was fitted to present effect estimates as risk ratios.ResultThe prevalence of spontaneous preterm birth was 14.5% (0.5% non-spontaneous). Characteristics not varying in pregnancy associated with increased risk of preterm birth were maternal age less than 18 years (adjusted risk ratio=1.13, 95% CI: 1.02 to 1.26); being Muslim (1.53, 1.16 to 2.01); first pregnancy (1.15, 1.04 to 1.28); multiple births (4.91, 4.20 to 5.75) and male child (1.10, 1.02 to 1.17). Those associated with decreased risk were maternal education >5 years (0.81, 0.73 to 0.90); maternal height ≥150 cm (0.89, 0.81 to 0.98) and being from wealthier families (0.83, 0.74 to 0.93). Pregnancy-related morbidities associated with increased risk of preterm birth were vaginal bleeding (1.53, 1.08 to 2.18); swelling (1.37, 1.17 to 1.60); high systolic blood pressure (BP) (1.47, 1.08 to 2.01) and high diastolic BP (1.41, 1.17 to 1.70) in the third trimester. Those associated with decreased risk were respiratory problem in the third trimester (0.86, 0.79 to 0.94); having poor appetite, nausea and vomiting in the second trimester (0.86, 0.80 to 0.92) and third trimester (0.86, 0.79 to 0.94); and higher weight gain from second to third trimester (0.89, 0.87 to 0.90).ConclusionThe prevalence of preterm birth is high in rural Nepal. Interventions that increase maternal education may play a role. Monitoring morbidities during antenatal care to intervene to reduce them through an effective health system may help reduce preterm birth.

COVID-19 has challenged health systems to learn how to learn. This paper describes the context, methods and challenges for learning to improve COVID-19 care at one academic health center. Challenges to learning include: (1) choosing a right clinical target; (2) designing methods for accurate predictions by borrowing strength from prior patients' experiences; (3) communicating the methodology to clinicians so they understand and trust it; (4) communicating the predictions to the patient at the moment of clinical decision; and (5) continuously evaluating and revising the methods so they adapt to changing patients and clinical demands. To illustrate these challenges, this paper contrasts two statistical modeling approaches - prospective longitudinal models in common use and retrospective analogues complementary in the COVID-19 context - for predicting future biomarker trajectories and major clinical events. The methods are applied to and validated on a cohort of 1,678 patients who were hospitalized with COVID-19 during the early months of the pandemic. We emphasize graphical tools to promote physician learning and inform clinical decision making., 16 pages, 7 figures, 1 table1

Age-related deficits in pattern separation have been postulated to bias the output of hippocampal memory processing toward pattern completion, which can cause deficits in accurate memory retrieval. Although the CA3 region of the hippocampus is often conceptualized as a homogeneous network involved in pattern completion, growing evidence demonstrates a functional gradient in CA3 along the transverse axis, as pattern-separated outputs (dominant in the more proximal CA3) transition to pattern-completed outputs (dominant in the more distal CA3). We examined the neural representations along the CA3 transverse axis in young (Y), aged memory-unimpaired (AU), and aged memory-impaired (AI) rats when different changes were made to the environment. Functional heterogeneity in CA3 was observed in Y and AU rats when the environmental similarity was high (altered cues or altered environment shapes in the same room), with more orthogonalized representations in proximal CA3 than in distal CA3. In contrast, AI rats showed reduced orthogonalization in proximal CA3 but showed normal (i.e., generalized) representations in distal CA3, with little evidence of a functional gradient. Under experimental conditions when the environmental similarity was low (different rooms), representations in proximal and distal CA3 remapped in all rats, showing that CA3 of AI rats is able to encode distinctive representations for inputs with greater dissimilarity. These experiments support the hypotheses that the age-related bias toward hippocampal pattern completion is due to the loss in AI rats of the normal transition from pattern separation to pattern completion along the CA3 transverse axis.

Subjective memory complaints are common after coronary artery bypass grafting (CABG), but previous studies have concluded that such symptoms are more closely associated with depressed mood than objective cognitive dysfunction. We compared the incidence of self-reported memory symptoms at 3 and 12 months after CABG with that of a control group of patients with comparable risk factors for coronary artery disease but without surgery.Patients undergoing CABG (n = 140) and a demographically similar nonsurgical control group with coronary artery disease (n = 92) were followed prospectively at 3 and 12 months. At each follow-up time, participants were asked about changes since the previous evaluation in areas of memory, calculations, reading, and personality. A Functional Status Questionnaire (FSQ) and self-report measure of symptoms of depression (CES-D) were also completed.The frequency of self-reported changes in memory, personality, and reading at 3 months was significantly higher among CABG patients than among nonsurgical controls. By contrast, there were no differences in the frequency of self-reported symptoms relating to calculations or overall rating of functional status. After adjustment for a measure of depression (CES-D rating score), the risk for self-reported memory changes remained nearly 5 times higher among the CABG patients than among control subjects. The relative risk of developing new self-reported memory symptoms between 3 and 12 months was 2.5 times higher among CABG patients than among nonsurgical controls (CI 1.24-5.02), and the overall prevalence of memory symptoms at 12 months was also higher among CABG patients (39%) than controls (14%).The frequency of self-reported memory symptoms 3 and 12 months after baseline is significantly higher among CABG patients than control patients with comparable risk factors for coronary and cerebrovascular disease. These differences could not be accounted for by symptoms of depression. The self-reported cognitive symptoms appear to be relatively specific for memory and may reflect aspects of memory functioning that are not captured by traditional measures of new verbal learning and memory. The etiology of these self-reported memory symptoms remains unclear, but our findings, as well as those of others, may implicate factors other than cardiopulmonary bypass itself.

A time series is a sequence of observations made over time. Examples in public health include daily ozone concentrations, weekly admissions to an emergency department or annual expenditures on health care in the United States. Time series models are used to describe the dependence of the response at each time on predictor variables including covariates and possibly previous values in the series. Time series methods are necessary to account for the correlation among repeated responses over time. This paper gives an overview of time series ideas and methods used in public health research.

Procedures for analyzing and comparing healthcare providers' effects on health services delivery and outcomes have been referred to as provider profiling. In a typical profiling procedure, patient-level responses are measured for clusters of patients treated by providers that in turn, can be regarded as statistically exchangeable. Thus, a hierarchical model naturally represents the structure of the data. When provider effects on multiple responses are profiled, a multivariate model rather than a series of univariate models, can capture associations among responses at both the provider and patient levels. When responses are in the form of charges for healthcare services and sampled patients include non-users of services, charge variables are a mix of zeros and highly-skewed positive values that present a modeling challenge. For analysis of regressor effects on charges for a single service, a frequently used approach is a two-part model (Duan, Manning, Morris, and Newhouse 1983) that combines logistic or probit regression on any use of the service and linear regression on the log of positive charges given use of the service. Here, we extend the two-part model to the case of charges for multiple services, using a log-linear model and a general multivariate log-normal model, and employ the resultant multivariate two-part model as the within-provider component of a hierarchical model. The log-linear likelihood is reparameterized as proposed by Fitzmaurice and Laird (1993), so that regressor effects on any use of each service are marginal with respect to any use of other services. The general multivariate log-normal likelihood is constructed in such a way that variances of log of positive charges for each service are provider-specific but correlations between log of positive charges for different services are uniform across providers. A data augmentation step is included in the Gibbs sampler used to fit the hierarchical model, in order to accommodate the fact that values of log of positive charges are undefined for unused service. We apply this hierarchical, multivariate, two-part model to analyze the effects of primary care physicians on their patients' annual charges for two services, primary care and specialty care. Along the way, we also demonstrate an approach for incorporating prior information about the effects of patient morbidity on response variables, to improve the accuracy of provider profiles that are based on patient samples of limited size.

Objective. To examine effects of primary care physicians (PCPs) and patients on the association between charges for primary care and specialty care in a point-of-service (POS) health plan.Data Source. Claims from 1996 for 3,308 adult male POS plan members, each of whom was assigned to one of the 50 family practitioner-PCPs with the largest POS plan member-loads.Study Design. A hierarchical multivariate two-part model was fitted using a Gibbs sampler to estimate PCPs' effects on patients' annual charges for two types of services, primary care and specialty care, the associations among PCPs' effects, and within-patient associations between charges for the two services. Adjusted Clinical Groups (ACGs) were used to adjust for case-mix. Principal Findings. PCPs with higher case-mix adjusted rates of specialist use were less likely to see their patients at least once during the year (estimated correlation: .40; 95% CI: .71, .008) and provided fewer services to patients that they saw (estimated correlation: .53; 95% CI: .77, .21). Ten of 11 PCPs whose case-mix adjusted effects on primary care charges were significantly less than or greater than zero (p < .05) had estimated, case-mix adjusted effects on specialty care charges that were of opposite sign (but not significantly different than zero). After adjustment for ACG and PCP effects, the within-patient, estimated odds ratio for any use of primary care given any use of specialty care was .57 (95% CI: .45, .73).Conclusions. PCPs and patients contributed independently to a trade-off between utilization of primary care and specialty care. The trade-off appeared to partially offset significant differences in the amount of care provided by PCPs. These findings were possible because we employed a hierarchical multivariate model rather than separate univariate models.

Background: The analysis and interpretation of change in cognitive function test scores after Coronary Artery Bypass Grafting (CABG). Longitudinal studies with multiple outcomes present considerable statistical challenges. Application of hierarchical linear statistical models can estimate the effects of a surgical intervention on the time course of multiple biomarkers. Methods: We use an "analyze then summarize" approach whereby we estimate the intervention effects separately for each cognitive test and then pool them, taking appropriate account of their statistical correlations. The model accounts for dropouts at follow-up, the chance of which may be related to past cognitive score, by implicitly imputing the missing data from individuals' past scores and group patterns.We apply this approach to a study of the effects of CABG on the time course of cognitive function as measured by 16 separate neuropsychological test scores, clustered into 8 cognitive domains. The study includes measurements on 140 CABG patients and 92 nonsurgical controls at baseline, and 3, 12, and 36 months. Including a nonsurgical control group allows comparison of changes in cognition over time between the surgery group and patients with similar risk factors, controlling for potential effects of aging and vascular disease.Results: We find that CABG patients have very longitudinal changes from baseline in cognitive function similar to those observed for nonsurgical controls. Any small differences tend to favor greater improvement in CABG patients than in the nonsurgical controls. Conclusions: The methods used have application to a wide range of intervention studies in which multiple biomarkers are followed over time to quantify health effects. Software to implement the methods in commonly used statistical packages is available from the authors at http://www.biostat.jhsph.edu/research/software.shtml.

Time series models relating short-term changes in air pollution levels to daily mortality counts typically assume that the effects of air pollution on the log relative rate of mortality do not vary with time. However, these short-term effects might plausibly vary by season. Changes in the sources of air pollution and meteorology can result in changes in characteristics of the air pollution mixture across seasons. The authors develop Bayesian semi-parametric hierarchical models for estimating time-varying effects of pollution on mortality in multi-site time series studies. The methods are applied to the updated National Morbidity and Mortality Air Pollution Study database for the period 1987--2000, which includes data for 100 U.S. cities. At the national level, a 10 micro-gram/m3 increase in PM(10) at lag 1 is associated with a 0.15 (95% posterior interval: -0.08, 0.39),0.14 (-0.14, 0.42), 0.36 (0.11, 0.61), and 0.14 (-0.06, 0.34) percent increase in mortality for winter, spring, summer, and fall, respectively. An analysis by geographical regions finds a strong seasonal pattern in the northeast (with a peak in summer) and little seasonal variation in the southern regions of the country. These results provide useful information for understanding particle toxicity and guiding future analyses of particle constituent data.

Patterned missing covariate data is a challenging issue in environmental epidemiology. For example, particulate matter measures of air pollution are often collected only every third day or every sixth day, while morbidity and mortality outcomes are collected daily. In this setting, many desirablemodels cannot be directly fit.We investigate such a setting in so-called “distributed lag” models when the lagged predictor is collected on a cruder time scale than the response. In multi-site studies with complete predictor data at some sites, multilevel models can be used to inform imputation for the sites with missing data. We focus on the implementation of such multilevel models, in terms of both modeldevelopment and computational implementation of the sampler. Specifically, we parallelize single chain runs of sampler. This is of note, since theMarkovian structure of Markov chain Monte Carlo (MCMC) samplers typically makes effective parallelization of single chains difficult. However, the conditional independence relationships of our developed model allow us to exploit parallel computing to run the chain. As a first attempt at using parallel MCMC for Bayesian imputation on such data, this chapter largely represents a proof of principle, though we demonstrate some promising potential for the methodology. Specifically, the methodology results in proportional decreases in run-time over the nonparallelized version near one over the number of available nodes. In addition, we describe a novel software implementation of parallelization that isuniquely suited to disk-based shared memory systems. We use a “blackboard” parallel computing scheme where shared network storage is a used as a blackboard to tally currently completed and queued tasks. This strategy allows for easy addition and subtraction of compute nodes and control of load balancing. Moreover, it builds in automatic checkpointing. Our investigation is motivated by multi-site time series studies of the short-term effectsof air pollution on disease or death rates. A commonmeasure of air pollution used for such studies is the amount in micrograms per cubic meter of particulate matter of a specified maximumaerodynamic diameter.We focus onPM2.5 (see Samet et al., 2000). Unfortunately, the definitive source of particulate matter data in the United States, the Environmental Protection Agency’s air pollution network of monitoring stations, collects data only a few times per week at some locations. One of the most frequent observed data patterns for2.5 are collected daily. In this setting, directly fitting a model that includes several lags of PM2.5 simultaneouslyis not possible. Such models are useful, for example, to investigate a cumulative weekly effect of air pollution on health. They are also useful tomore finely investigate the dynamics of the relationship between the exposure and response.As an example, onemight postulate that after an increase in air pollution, high air pollution levels on later days may have a smaller impact, as the risk set has been depleted from the initial increase (Dominici et al., 2002; Schwartz, 2000; Zeger et al., 1999). We focus on distributed lag models that relate the current-day disease rate to particulatematter levels over the past week. That is, our model includes the current day’s PM2.5 levels as well as the previous six days. While direct estimation of the effect for any particular lag is possible, joint estimation of the distributed lag model is not possible (see Section 20.3). Moreover, missing-data imputation for counties with patterned missing data is difficult. We consider a situation where several independent time series are observed at different geographical regions, somewith complete PM2.5 data. We usemultilevel models to borrow information across series tofill in themissingdata viaBayesian imputation. Thehierarchical model is also used to combine county-specific distributed lag effects into national estimates. The rest of the chapter is organized as follows. In Section 20.2 we outline the data setused for analysis and follow in Section 20.3 with a discussion of Bayesian imputation. In Section 20.4 we describe the distributed lag models of interest, and in Section 20.5 we illustrate a multiple imputation strategy. Section 20.6 uses the imputation algorithm to analyze hospitalization rates of chronic obstructive pulmnonary disease (COPD). Finally, Section 20.7 gives some conclusions, discussion and proposals for future work.