Dataset Information

Large numbers of explanatory variables, a semi-descriptive analysis.

ABSTRACT: Data with a relatively small number of study individuals and a very large number of potential explanatory features arise particularly, but by no means only, in genomics. A powerful method of analysis, the lasso [Tibshirani R (1996) J Roy Stat Soc B 58:267-288], takes account of an assumed sparsity of effects, that is, that most of the features are nugatory. Standard criteria for model fitting, such as the method of least squares, are modified by imposing a penalty for each explanatory variable used. There results a single model, leaving open the possibility that other sparse choices of explanatory features fit virtually equally well. The method suggested in this paper aims to specify simple models that are essentially equally effective, leaving detailed interpretation to the specifics of the particular study. The method hinges on the ability to make initially a very large number of separate analyses, allowing each explanatory feature to be assessed in combination with many other such features. Further stages allow the assessment of more complex patterns such as nonlinear and interactive dependences. The method has formal similarities to so-called partially balanced incomplete block designs introduced 80 years ago [Yates F (1936) J Agric Sci 26:424-455] for the study of large-scale plant breeding trials. The emphasis in this paper is strongly on exploratory analysis; the more formal statistical properties obtained under idealized assumptions will be reported separately.

SUBMITTER: Cox DR

PROVIDER: S-EPMC5559019 | biostudies-literature | 2017 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Large numbers of explanatory variables, a semi-descriptive analysis.

Cox D R DR Battey H S HS

Proceedings of the National Academy of Sciences of the United States of America 20170724 32

Data with a relatively small number of study individuals and a very large number of potential explanatory features arise particularly, but by no means only, in genomics. A powerful method of analysis, the lasso [Tibshirani R (1996) <i>J Roy Stat Soc B</i> 58:267-288], takes account of an assumed sparsity of effects, that is, that most of the features are nugatory. Standard criteria for model fitting, such as the method of least squares, are modified by imposing a penalty for each explanatory var ...[more]

PMID: 28739925

Similar Datasets

Project description:BackgroundThe Surgical Risk Preoperative Assessment System (SURPAS) is a parsimonious set of models providing accurate preoperative prediction of common adverse outcomes for individual patients. However, focus groups with surgeons and patients have developed a list of questions about and recommendations for how to further improve SURPAS's usability and usefulness. Eight issues were systematically evaluated to improve SURPAS.MethodsThe eight issues were divided into three groups: concerns to be addressed through further analysis of the database; addition of features to the SURPAS tool; and the collection of additional outcomes. Standard multiple logistic regression analysis was performed using the 2005-2015 American College of Surgeons National Surgical Quality Improvement Participant Use File (ACS NSQIP PUF) to refine models: substitution of the preoperative sepsis variable with a procedure-related risk variable; testing of an indicator variable for multiple concurrent procedure codes in complex operations; and addition of outcomes to increase clinical applicability. Automated risk documentation in the electronic health record and a patient handout and supporting documentation were developed. Long term functional outcomes were considered.ResultsModel discrimination and calibration improved when preoperative sepsis was replaced with a procedure-related risk variable. Addition of an indicator variable for multiple concurrent procedures did not significantly improve the models. Models were developed for a revised set of eleven adverse postoperative outcomes that separated bleeding/transfusion from the cardiac outcomes, UTI from the other infection outcomes, and added a predictive model for unplanned readmission. Automated documentation of risk assessment in the electronic health record, visual displays of risk for providers and patients and an "About" section describing the development of the tool were developed and implemented. Long term functional outcomes were considered to be beyond the scope of the current SURPAS tool.ConclusionRefinements to SURPAS were successful in improving the accuracy of the models, while reducing manual entry to five of the eight variables. Adding a predictor variable to indicate a complex operation with multiple current procedure codes did not improve the accuracy of the models. We developed graphical displays of risk for providers and patients, including a take-home handout and automated documentation of risk in the electronic health record. These improvements should facilitate easier implementation of SURPAS.

Project description:BACKGROUND: The performance of Primary Care Trusts in England is assessed and published using a number of different performance indicators. Our study has two broad purposes. Firstly, to find out whether pairs of indicators that purport to measure similar aspects of quality are correlated (as would be expected if they are both valid measures of the same construct). Secondly, we wanted to find out whether broad (global) indicators correlated with any particular features of Primary Care Trusts, such as expenditure per capita. METHODS: Cross sectional quantitative analysis using data from six 2004/05 PCT performance indicators for 303 English Primary Care Trusts from four sources in the public domain: Star Rating, aggregated Quality and Outcomes Framework scores, Dr Foster mortality index, Dr Foster equity index (heart by-pass and hip replacements), NHS Litigation Authority Risk Management standards and Patient Satisfaction scores from the Star Ratings. Forward stepwise multiple regression analysis to determine the effect of Primary Care Trust characteristics on performance. RESULTS: Star Rating and Quality and Outcomes Framework total, both summary measures of global quality, were not correlated with each other (F = 0.66, p = 0.57). There were however positive correlations between Quality and Outcomes Framework total and patient satisfaction (r = 0.61, p < 0.001) and between screening/'additional services' indicators on the Star Ratings and Quality and Outcomes Framework (F = 24, p < 0.001). There was no correlation between different measures of access to services. Likewise we found no relationship between either Star Rating or Litigation Authority Standards and hospital mortality (F = 0.61, p = 0.61; F = 0.31, p = 0.73). CONCLUSION: Performance assessment in healthcare remains on the Government's agenda, with new core and developmental standards set to replace the Star Ratings in 2006. Yet the results of this analysis provide little evidence that the current indicators have sufficient construct validity to measure the underlying concept of quality, except when the specific area of screening is considered.

Dataset Information

Large numbers of explanatory variables, a semi-descriptive analysis.

Publications

Large numbers of explanatory variables, a semi-descriptive analysis.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets