Abstract:Patient diagnosis and treatment involves factors that occur at different times. In order to identify as early as possible those factors associated with excessive length of stay in a hospital, time staging was added to induced decision trees. Time-staged induced decision trees uncovered useful patterns in data from the earliest time period, which represents a patient’s medical history, even though linear regression did not identify any significant variables in this data. Introduction Induced decision trees have been successfully applied to problems in medical diagnosis, both in identifying an illness (Kononenko et al. 1984; Mingers 1989) and in determining the most cost-effective diagnostic tests (Nunez 1991). Induction operates on a set of cases with known outcomes, along with their associated symptoms or diagnostic tests. Induction methods, such as induced decision trees and linear regression identify those attributes most significant in predicting the outcome. For induced decision trees, also called rule induction, the output takes the form of a decision tree. The most significant attribute is located at the root of the tree; succeeding attributes further discriminate between the outcomes. The sequence of attribute values in the decision tree can easily be converted to the rules of an expert system. The explanatory power of these rules is one of the strong points of induced decision trees. Another strength is their ability, if constructed in stages, to show data attributes in the order in which they occur (Murphy and Benaroch 1997). Alternate modes of predicting outcomes include neural networks. Although neural networks lack explanatory power, they are the most accurate predictors, in most studies. Modeling data in time stages is a useful approach for studying the diagnosis and treatment of diseases as they progress. The application discussed in this paper is identifying critical factors for patients undergoing knee replacement surgery. The tertiary care hospital that provided the data was implementing critical pathways, standard clinical steps in treating a disease. Their goal is reducing costs and improving quality. The data collected on a hospitalized patient is available in different time frames. The medical history is available before admission. Other data, such as the results of tests, and, in the case of knee replacement, the distance walked become available at varying points during the hospital stay. The goal of this research is early identification of factors that affect length of stay in order to implement interventions that address the problems. The resulting induced decision trees present data in its natural order of availability. The trees uncover several factors contributing to excessive length of stay. Experimental Method Data was collected on 29 attributes for 119 patients who had knee replacement surgery. The conventional rule induction algorithm was modified so that it used data in its order of availability. The modified attribute selection step forced any statistically significant attributes from the first time period to enter the tree before attributes from the next time period. Later time groups entered, one by one, when no attributes from the prior groups were significant. Results Analysis with linear regression identified only one significant attribute, the distance walked on the fourth day; however, the induced decision trees revealed useful patterns in data from the earliest time period. The trees can uncover those early attributes that act in conjunction to influence length of stay. For example, the tree in Figure 1 contributes several rules; one is: if patients do not use a cane prior to admission and have a body mass (pounds/inches-squared) less than .05, they have an 80% chance of leaving in 8 days. 45 patients with this profile were discharged in 8 days or less; 11 stayed longer. (A body mass of .05 corresponds to a person 5’6” weighing 218 pounds.) Another rule in the tree is that if patients are able to walk more than 5 feet on day 3 but cannot walk on day 4, they will stay more than 8 days in the hospital. This rule agrees with comments that appear ed in patient records. When patients did well on their first walk in physical therapy, but not their second, their condition had deteriorated. They felt dizzy, weak, or in pain.

Enzyme-free Colorimetric Determination of Uric Acid Based on Inhibition of Gold Nanorods Etching

Enhanced Enzyme-Like DNAzyme@MOFs for Non-Enzymatic Uric Acid Detection in Interstitial Fluid

Colorimetric and visual determination of uric acid based on decolorization of manganese dioxide nanosheet dispersions

A Nanostructured Sensor Based on Gold Nanoparticles and Nafion for Determination of Uric Acid

A Biosensor Based on the Biomimetic Oxidase Fe3O4@MnO2 for Colorimetric Determination of Uric Acid

Enzyme-free and wide-range portable colorimetric sensing system for uric acid and hydrogen peroxide based on copper nanoparticles

Simple enzyme-free detection of uric acid by an in situ fluorescence and colorimetric method based on Co-PBA with high oxidase activity

Target-triggered 'colorimetric-fluorescence' dual-signal sensing system based on the versatility of MnO 2 nanosheets for rapid detection of uric acid

A microfluidic paper-based colorimetric device for the visual detection of uric acid in human urine samples

Americas Conference on Information Systems ( AMCIS ) December 1998 Induced Decision Trees for Temporal Medical Data

A ''naked-eye'' colorimetric and ratiometric fluorescence probe for uric acid based on Ti<sub>3</sub>C<sub>2</sub> MXene quantum dots

Continual and accurate home monitoring of uric acid in urine samples with uricase-packaged nanoflowers assisted portable electrochemical uricometer

Electrochemical Enhanced Detection of Uric Acid Based on Peroxidase‐like Activity of Fe 3 O 4 @Au

Diagnosing pulmonary alveolar proteinosis. A review and an update.

A Novel Uric Acid Biosensor Based on Regular Prussian Blue Nanocrystal/Upright Graphene Oxide Array Nanocomposites

Enhanced peroxidase‐like activity of MnFe2O4 nanoparticles on halloysite nanotubes for uric acid detection

Bimetallic Nanoparticles Decorated Hollow Nanoporous Carbon Framework As Nanozyme Biosensor for Highly Sensitive Electrochemical Sensing of Uric Acid.

Reliable quantitative detection of uric acid in urine by surface-enhanced Raman spectroscopy with endogenous internal standard

Sensitive and Highly Selective Biosensor Based on Triangular Au Nanoplates for Detection of Uric Acid in Human Serum

Uric acid quantification via colorimetric detection utilizing silver oxide-modified activated carbon nanoparticles functionalized with ionic liquid

What Do People Like to “Share” About Obesity? A Content Analysis of Frequent Retweets About Obesity on Twitter