Rethinking PICO in the Machine Learning Era: ML-PICO

Background  Machine learning (ML) has captured the attention of many clinicians who may not have formal training in this area but are otherwise increasingly exposed to ML literature that may be relevant to their clinical specialties. ML papers that follow an outcomes-based research format can be assessed using clinical research appraisal frameworks such as PICO (Population, Intervention, Comparison, Outcome). However, the PICO frameworks strain when applied to ML papers that create new ML models, which are akin to diagnostic tests. There is a need for a new framework to help assess such papers. Objective  We propose a new framework to help clinicians systematically read and evaluate medical ML papers whose aim is to create a new ML model: ML-PICO (Machine Learning, Population, Identification, Crosscheck, Outcomes). We describe how the ML-PICO framework can be applied toward appraising literature describing ML models for health care. Conclusion  The relevance of ML to practitioners of clinical medicine is steadily increasing with a growing body of literature. Therefore, it is increasingly important for clinicians to be familiar with how to assess and best utilize these tools. In this paper we have described a practical framework on how to read ML papers that create a new ML model (or diagnostic test): ML-PICO. We hope that this can be used by clinicians to better evaluate the quality and utility of ML papers.     

Electronic health records and support for primary care teamwork

Objective Consensus that enhanced teamwork is necessary for efficient and effective primary care delivery is growing. We sought to identify how electronic health records (EHRs) facilitate and pose challenges to primary care teams as well as how practices are overcoming these challenges.Methods Practices in this qualitative study were selected from those recognized as patient-centered medical homes via the National Committee for Quality Assurance 2011 tool, which included a section on practice teamwork. We interviewed 63 respondents, ranging from physicians to front-desk staff, from 27 primary care practices ranging in size, type, geography, and population size.Results EHRs were found to facilitate communication and task delegation in primary care teams through instant messaging, task management software, and the ability to create evidence-based templates for symptom-specific data collection from patients by medical assistants and nurses (which can offload work from physicians). Areas where respondents felt that electronic medical record EHR functionalities were weakest and posed challenges to teamwork included the lack of integrated care manager software and care plans in EHRs, poor practice registry functionality and interoperability, and inadequate ease of tracking patient data in the EHR over time.Discussion Practices developed solutions for some of the challenges they faced when attempting to use EHRs to support teamwork but wanted more permanent vendor and policy solutions for other challenges.Conclusions EHR vendors in the United States need to work alongside practicing primary care teams to create more clinically useful EHRs that support dynamic care plans, integrated care management software, more functional and interoperable practice registries, and greater ease of data tracking over time.     

Healing the Professional Culture of Medicine

The past decade has been a time of great change for US physicians. Many physicians feel that the care delivery system has become a barrier to providing high-quality care rather than facilitating it. Although physician distress and some of the contributing factors are now widely recognized, much of the distress physicians are experiencing is related to insidious issues affecting the cultures of our profession, our health care organizations, and the health care delivery system. Culture refers to the shared and fundamental beliefs of a group that are so widely accepted that they are implicit and often no longer recognized. When challenges with culture arise, they almost always relate to a problem with a subcomponent of the culture even as the larger culture does many things well. In this perspective, we consider the role of culture in many of the problems facing our health care delivery system and contributing to the high prevalence of professional burnout plaguing US physicians. A framework, drawn from the field of organizational science, to address these issues and heal our professional culture is considered.     

Harnessing Behavioral Economics Principles to Promote Better Surgeon Accountability for Operating Room Cost: A Prospective Study

1. Download : Download high-res image (295KB)
2. Download : Download full-size image

     

Effect of maternal and infant covariates on sibship correlation in birth weight

Birth weight on 12,644 singleton infants from 6,196 sibships born in Maryland between 1980 and 1984 were used to estimate the effects of nine maternal and infant covariates on the sibship correlation in birth weight. Assuming a homogeneous correlation across all families, the estimated intraclass correlation was 0.4664 (+/- 0.0099). This high sibship correlation makes it possible to predict, with reasonable accuracy, the birth weight of a child given information on previous sibs, as well as covariates on the mother and/or infant pertinent to a given pregnancy. The reduction in variance associated with incorporating information on the nine covariates used here was approximately equal to that obtained by conditioning on a single previous sib. Testing for heterogeneity in correlation among different groups of families showed that a crude measure of parity (first live birth vs. other), time between births, mother's marital status, and maternal age at the birth of the last child significantly influenced the sibship correlation in birth weight.     

Etiological heterogeneity in Hodgkin's disease: HLA linked and unlinked determinants of susceptibility independent of histological concordance

Forty-one multiplex families, from published sources and new data from the National Cancer Institute, segregating for Hodgkin's disease and HLA, have been studied. A reanalysis of these data strongly suggests a recessive mode of inheritance for susceptibility to Hodgkin's disease. The HLA haplotype sharing data between affected relatives demonstrate that approximately 60% of cases in multiplex families are due to an HLA-linked susceptibility gene, the remaining 40% being due to other familial factors. The data clearly support the hypothesis of etiological heterogeneity for Hodgkin's disease, with both HLA-linked and HLA-unlinked factors being responsible. Finally, there is an increased concordance of histological types between affected relatives, but this concordance seems independent of HLA sharing.     

Optimal Selection of Sib Pairs from Random Samples for Linkage Analysis of a QTL Using the EDAC Test

Percentages of extremely concordant and extremely discordant sib pairs are calculated that maximize the power to detect a quantitative trait locus (QTL) under a variety of circumstances using the EDAC test. We assume a large fixed number of randomly sampled sib pairs, such as one would hope to find in the large twin registries, and limited resources to genotype a certain number of selected sib pairs. Our aim is to investigate whether optimal selection can be achieved when prior knowledge concerning the QTL gene action, QTL allele frequency, QTL effect size, and background (residual) sib correlation is limited or absent. To this end we calculate the best selection percentages for a large number of models, which differ in QTL gene action allele frequency, background correlation, and QTL effect size. By averaging these percentages over gene action, over allele frequency, over gene action, and over allele frequencies, we arrive at general recommendations concerning selection percentages. The soundness of these recommendations is subsequently in a number of test cases.     

Path analysis and sib-pair linkage

When the distance between linked loci is expressed in terms of the correlation between the identity-by-descent (idb) values of the loci, then a path model may be used to order loci with data on sib-pairs and their parents. The relationship between the recombination fraction and the correlation coefficient is developed and a method for fitting a covariance matrix predicted by a specific ordering of loci to an observed covariance matrix is proposed. © 1993 Wiley-Liss. Inc.