Model selection in competing risks regression

In the analysis of time‐to‐event data, the problem of competing risks occurs when an individual may experience one, and only one, of m different types of events. The presence of competing risks complicates the analysis of time‐to‐event data, and standard survival analysis techniques such as Kaplan–Meier estimation, log‐rank test and Cox modeling are not always appropriate and should be applied with caution. Fine and Gray developed a method for regression analysis that models the hazard that corresponds to the cumulative incidence function. This model is becoming widely used by clinical researchers and is now available in all the major software environments. Although model selection methods for Cox proportional hazards models have been developed, few methods exist for competing risks data. We have developed stepwise regression procedures, both forward and backward, based on AIC, BIC, and BICcr (a newly proposed criteria that is a modified BIC for competing risks data subject to right censoring) as selection criteria for the Fine and Gray model. We evaluated the performance of these model selection procedures in a large simulation study and found them to perform well. We also applied our procedures to assess the importance of bone mineral density in predicting the absolute risk of hip fracture in the Women's Health Initiative–Observational Study, where mortality was the competing risk. We have implemented our method as a freely available R package called crrstep. Copyright © 2013 John Wiley & Sons, Ltd.     

The performance of automated case-mix adjustment regression model building methods in a health outcome prediction setting

We have previously described a system for monitoring a number of healthcare outcomes using case-mix adjustment models. It is desirable to automate the model fitting process in such a system if monitoring covers a large number of outcome measures or subgroup analyses. Our aim was to compare the performance of three different variable selection strategies: “manual”, “automated” backward elimination and re-categorisation, and including all variables at once, irrespective of their apparent importance, with automated re-categorisation. Logistic regression models for predicting in-hospital mortality and emergency readmission within 28 days were fitted to an administrative database for 78 diagnosis groups and 126 procedures from 1996 to 2006 for National Health Services hospital trusts in England. The performance of models was assessed with Receiver Operating Characteristic (ROC) c statistics, (measuring discrimination) and Brier score (assessing the average of the predictive accuracy). Overall, discrimination was similar for diagnoses and procedures and consistently better for mortality than for emergency readmission. Brier scores were generally low overall (showing higher accuracy) and were lower for procedures than diagnoses, with a few exceptions for emergency readmission within 28 days. Among the three variable selection strategies, the automated procedure had similar performance to the manual method in almost all cases except low-risk groups with few outcome events. For the rapid generation of multiple case-mix models we suggest applying automated modelling to reduce the time required, in particular when examining different outcomes of large numbers of procedures and diseases in routinely collected administrative health data.     

CRITIQUE OF OCCUPATIONAL HEALTH MONITORING IN AUSTRALIA

The procedures used to collect data relating to occupational health in Australia are reviewed and a critique is offered of their adequacy in presenting a comprehensive picture of occupational health. Attention is focused on the death certification process and on the role of the cancer and perinatal statistics units. Proposals are offered for overhauling the social procedures for collecting public health data, with a view to providing a better base for preventive strategies.     

Analysis of incomplete public health data

The problem of dealing with missing values is common throughout statistics and is very prominent with epidemiologic data in the broad sense. Not only do data collection procedures break down, but subjects may be lost to follow up, or simply withdraw their consent without further providing a reason for doing so. In this paper, we review a framework for handling incomplete studies, and then concentrate on a specific case. It comes from a complex health interview survey, conducted in Belgium in 1997, where different types of missingness arise at various levels of the hierarchical sampling procedure.     

Penalized regression procedures for variable selection in the potential outcomes framework

A recent topic of much interest in causal inference is model selection. In this article, we describe a framework in which to consider penalized regression approaches to variable selection for causal effects. The framework leads to a simple ‘impute, then select’ class of procedures that is agnostic to the type of imputation algorithm as well as penalized regression used. It also clarifies how model selection involves a multivariate regression model for causal inference problems and that these methods can be applied for identifying subgroups in which treatment effects are homogeneous. Analogies and links with the literature on machine learning methods, missing data, and imputation are drawn. A difference least absolute shrinkage and selection operator algorithm is defined, along with its multiple imputation analogs. The procedures are illustrated using a well‐known right‐heart catheterization dataset. Copyright © 2015 John Wiley & Sons, Ltd.     

Routine documentation based on disability data of the legal health insurance

Data on sickness absence of employees are routinely used for health reporting and the analysis of work-related morbidity by institutions of the German health insurance. Since the insurance system comprises several different branches, these health reports differ in respect of objectives, data selection, methods of analysis, and presentation of results. A further lack of comparability is caused by the heterogeneous populations, since membership in a certain health insurance depends on social status and job requirements. Aim of this paper was to review the methods and characteristics of health reports as they are routinely published in Germany. By evaluating these reports recommendations should be derived to improve comparability. The review showed that at least four different kinds of health reports should be differentiated: company-based health reports, health reports for business branches, morbidity statistics, and reports from research projects. Such reports have different objectives and therefore require different methods of data analysis. However, group-specific and common standards could be set up and it is suggested that health insurance institutions should work out report guidelines based on the recommendations given in this review.     

An exploration of medical professionals' perceptions of managed care and policy concepts: a magnitude estimation approach

The authors begin by establishing the information content of measurement as the critical element in seeking research findings. They go on to indicate that traditional social science scaling has not provided all the information necessary to accomplish the task of measurement. They review some of the relevant literature in the field where the most common measure, Likert type scaling, has been used. They contrast that literature with research based on magnitude estimation procedures. They then apply magnitude estimation procedures to study beliefs among physicians and health care providers regarding managed care. This procedure is a novel application of magnitude estimation having never been applied in managed care research. The magnitude estimation based data are determined to yield ratio scales with true zero points, thereby, representing the highest level of measurement information. The differences between the pairs of managed care concepts using these ratio scales show that the health care concepts have greater variation in distances between them than might have heretofore been recognized because of the limitations of the scales previously applied. Consequently, the data reported have the potential to yield greater information concerning the current state of physicians' and providers' beliefs regarding managed care. Policy implications are discussed.     

Comparison of two-phase analyses for case-control genetic association studies

To test for genetic association between a marker and a complex disease using a case-control design, Cochran-Armitage trend tests (CATTs) and Pearson's chi-square test are often employed. Both tests are genotype-based. Song and Elston (Statist. Med. 2006; 25:105-126) introduced the Hardy-Weinberg disequilibrium trend test and combined it with CATT to test for association. Compared to using a single statistic to test for case-control genetic association (referred to as single-phase analysis), two-phase analysis is a new strategy in that it employs two test statistics in one analysis framework, each statistic using all available case-control data. Two such two-phase analysis procedures were studied, in which Hardy-Weinberg equilibrium (HWE) in the population is a key assumption, although the procedures are robust to moderate departure from HWE. Our goal in this article is to study a new two-phase procedure and compare all three two-phase analyses and common single-phase procedures by extensive simulation studies. For illustration, the results are applied to real data from two case-control studies. On the basis of the results, we conclude that with an appropriate choice of significance level for the analysis in phase 1, some two-phase analyses could be more powerful than commonly used test statistics.     

Using Grounded Theory Method to Capture and Analyze Health Care Experiences

ObjectiveGrounded theory (GT) is an established qualitative research method, but few papers have encapsulated the benefits, limits, and basic tenets of doing GT research on user and provider experiences of health care services. GT can be used to guide the entire study method, or it can be applied at the data analysis stage only.MethodsWe summarize key components of GT and common GT procedures used by qualitative researchers in health care research. We draw on our experience of conducting a GT study on amyotrophic lateral sclerosis patients’ experiences of health care services.FindingsWe discuss why some approaches in GT research may work better than others, particularly when the focus of study is hard-to-reach population groups. We highlight the flexibility of procedures in GT to build theory about how people engage with health care services.ConclusionGT enables researchers to capture and understand health care experiences. GT methods are particularly valuable when the topic of interest has not previously been studied. GT can be applied to bring structure and rigor to the analysis of qualitative data.     

A robust approach for skewed and heavy-tailed outcomes in the analysis of health care expenditures

In this paper robust statistical procedures are presented for the analysis of skewed and heavy-tailed outcomes as they typically occur in health care data. The new estimators and test statistics are extensions of classical maximum likelihood techniques for generalized linear models. In contrast to their classical counterparts, the new robust techniques show lower variability and excellent efficiency properties in the presence of small deviations from the assumed model, i.e. when the underlying distribution of the data lies in a neighborhood of the model. A simulation study, an analysis on real data, and a sensitivity analysis confirm the good theoretical statistical properties of the new techniques.     

Rapid model exploration for complex hierarchical data: application to pharmacokinetics of insulin aspart

We consider situations, which are common in medical statistics, where we have a number of sets of response data, from different individuals, say, potentially under different conditions. A parametric model is defined for each set of data, giving rise to a set of random effects. Our goal here is to efficiently explore a range of possible ‘population’ models for the random effects, to select the most appropriate model. The range of possible models is potentially vast, because the random effects may depend on observed covariates, and there may be multiple credible ways of partitioning their variability. Here, we consider pharmacokinetic (PK) data on insulin aspart, a fast acting insulin analogue used in the treatment of diabetes. PK models are typically nonlinear (in their parameters), often complex and sometimes only available as a set of differential equations, with no closed‐form solution. Fitting such a model for just a single individual can be a challenging task. Fitting a joint model for all individuals can be even harder, even without the complication of an overarching model selection objective. We describe a two‐stage approach that decouples the population model for the random effects from the PK model applied to the response data but nevertheless fits the full, joint, hierarchical model, accounting fully for uncertainty. This allows us to repeatedly reuse results from a single analysis of the response data to explore various population models for the random effects. This greatly expedites not only model exploration but also cross‐validation for the purposes of model criticism. © 2015 The Authors. Statistics in Medicine published by John Wiley & Sons Ltd.     

Using threshold regression to analyze survival data from complex surveys: With application to mortality linked NHANES III Phase II genetic data

The Cox proportional hazards (PH) model is a common statistical technique used for analyzing time‐to‐event data. The assumption of PH, however, is not always appropriate in real applications. In cases where the assumption is not tenable, threshold regression (TR) and other survival methods, which do not require the PH assumption, are available and widely used. These alternative methods generally assume that the study data constitute simple random samples. In particular, TR has not been studied in the setting of complex surveys that involve (1) differential selection probabilities of study subjects and (2) intracluster correlations induced by multistage cluster sampling. In this paper, we extend TR procedures to account for complex sampling designs. The pseudo‐maximum likelihood estimation technique is applied to estimate the TR model parameters. Computationally efficient Taylor linearization variance estimators that consider both the intracluster correlation and the differential selection probabilities are developed. The proposed methods are evaluated by using simulation experiments with various complex designs and illustrated empirically by using mortality‐linked Third National Health and Nutrition Examination Survey Phase II genetic data.     

Using risk assessment to evaluate adverse selection under capitated contracts

Most provider organizations rely on health plan or market-based information about capitation rates, per member per month costs, and utilization trends to benchmark their performance. However, these statistics can be misleading because of differences in enrollee mix and contracting terms across provider organizations. This article describes the limitations of health plan contract provisions in protecting against adverse selection. It describes various actuarial and statistical data sources for evaluation of adverse selection. The article then presents various approaches to risk adjustment on population basis and their use in quantifying adverse selection for health plan contract negotiations.     

An orientation toward help-seeking for emotional problems

In recent years, many researchers tried to explain the social selection in use of mental health care services. A modest role is attributed to the orientation toward help-seeking. This article studies this orientation. Our research-population consisted of 10,171 Dutch persons, aged 18 and older. Analysis showed that most people are prone to seek help for one or more emotional problems. People who are more prone to seek help are younger, have had more education and have a higher family income. They have more often acquaintances working in mental health care. People who are more prone to seek help do not see chance as the locus of control of health. These people are less dependent on their GP for common disorders and are more open about mental health matters. The results of discriminant analysis are not satisfactory, but when we try to distinguish the groups of people who are and who are not willing to seek help, we see that the best discriminating factor is their help-seeking attitude for common disorders. People who have high expectations from the GP for common disorders, clearly do have a preference to seek help for the emotional problems. The groups of people who are more willing to seek help from the GP compared to mental health professionals cannot be distinguished by these expectations. Here the level of education discriminates fairly well: people who are more prone to seek help from a GP have a lower educational level. Future research should be focussed on the testing of a theoretical model that explains the orientation toward help-seeking for emotional problems and selection in help-seeking with longitudinal data.     

Predictability and predictiveness in health care spending

This paper re-examines the relation between the predictability of health care spending and incentives due to adverse selection. Within an explicit model of health plan decisions about service levels, we show that predictability (how well spending on certain services can be anticipated), predictiveness (how well the predicted levels of certain services contemporaneously co-vary with total health care spending), and demand responsiveness all matter for adverse selection incentives. The product of terms involving these three measures of predictability, predictiveness, and demand responsiveness define an empirical index of the direction and magnitude of selection incentives. We quantify the relative magnitude of adverse selection incentives bearing on various types of health care services in Medicare. Our results are consistent with other research on service-level selection. The index of incentives can readily be applied to data from other payers.     

统计软件配伍组秩和检验及多重比较

目的：针对医学研究中常见配伍组设计等级资料数据分析工作存在的一些问题，给予统计软件技术上正确且实用的支持。方法：灵活结合秩变换理论和软件本身的一些特点，分别在统计软件SAS、SPSS以及Stata中实现配伍组设计秩和检验及其多重比较。结果：3种软件对同一数据资料进行处理，所得结论基本相同。结论：本文提供了统计软件实现该种数据分析的详细的程序、具体的操作步骤和结果解释，相关研究工作和数据分析人员可以结合不同的工作条件和个人喜好，应用不同的软件正确地完成此类数据的分析工作。     

Monitoring health care system performance in Iowa.

Monitoring the performance of the health care delivery system is a public health function that becomes more important as organized delivery systems begin to take control over larger portions of the medical care market. The study reported here illustrates how standard medical care epidemiology can be applied to analysis of health care system performance to aid governmental efforts to monitor new developments in the medical care market. In order to evaluate the efficiency of hospital care delivered in Iowa, age- and sex-adjusted population admission rates for five common procedures were generated for all 99 counties. The five common procedures were defined as follows: hernia, tonsillectomy, cesarean section, hysterectomy, and cholecystectomy. In addition, variations in 11 ambulatory care-sensitive condition rates were analyzed. Residents of at least 15 counties were at significantly greater than average risk of receiving each of the common procedures (chi 2 test, P < 0.05). Counties that had a high rate for one procedure tended to have a high rate for at least one other procedure. Several counties had more than twice the mean rate. Even a 10% reduction, when added across all five common procedures. amounts to well over 2,000 hospitalizations avoided. It is assumed that reductions would be concentrated in high-rate counties. If a 50% reduction could be achieved in only part of the ambulatory care-sensitive procedures, more than 10,000 hospital admissions could be avoided.     

Haplotypes and haplotype-tagging single-nucleotide polymorphism: presentation Group 8 of Genetic Analysis Workshop 14

Moderately dense maps of single-nucleotide polymorphism (SNP) markers across the human genome for both the simulated data set and data from the Collaborative Study of the Genetics of Alcoholism were available at Genetic Analysis Workshop 14 for the first time. This allowed examination of various novel and existing methods for haplotype analyses. Three contributors applied Mantel statistics in different ways for both linkage and association analysis by using the shared length between two haplotypes at a marker locus as a measure of genetic similarity. The results indicate that haplotype-sharing based on Mantel statistics can be a powerful approach and needs further methodological evaluation. Four contributors investigated haplotype-tagging SNP (htSNP) selection procedures, two contributors examined the use of multilocus haplotypes compared to single loci in association tests, and two contributors compared the accuracy of various methods for reconstructing haplotypes and estimating haplotype frequencies for both pedigree data and data from unrelated individuals. For all three different tasks, software packages and procedures gave similar results in regions of high linkage disequilibrium (LD). However, they were not as consistent in regions of moderate to low LD. One coalescence-based approach for estimating haplotype frequencies, coupled with a Markov chain Monte Carlo technique, outperformed the other haplotype frequency estimation methods in regions of low LD. In conclusion, regardless of the task, results were similar in chromosomal regions of high LD. However, based on the differing results observed here, methodological improvements are required for chromosomal regions of low to moderate LD.     

Comparing hospital discharge records with death certificates: Can the differences be explained?

STUDY OBJECTIVE: The quality of mortality statistics is important for epidemiological research. Considerable discrepancies have been reported between death certificates and corresponding hospital discharge records. This study examines whether differences between the death certificate's underlying cause of death and the main condition from the final hospital discharge record can be explained by differences in ICD selection procedures. The authors also discuss the implications of unexplained differences for mortality data quality. DESIGN: Using ACME, a standard software for the selection of underlying cause of death, the compatibility between the underlying cause of death and the final main condition was examined. The study also investigates whether data available in the hospital discharge record, but not reported on the death certificate, influence the selection of the underlying cause of death. SETTING: Swedish death certificates for 1995 were linked to the national hospital discharge register. The resulting database comprised 69 818 people who had been hospitalised during their final year of life. MAIN RESULTS: The underlying cause of death and the main condition differed at Basic Tabulation List level in 54% of the deaths. One third of the differences could not be explained by ICD selection procedures. Adding hospital discharge data changed the underlying cause in 11% of deaths. For some causes of death, including medical misadventures and accidental falls, the effect was substantial. CONCLUSION: Most differences between underlying cause of death and final main condition can be explained by differences in ICD selection procedures. Further research is needed to investigate whether unexplained differences indicate lower data quality.