Evaluation Of Classification Methods For The Prediction Of Hospital Length

1320 Words6 Pages
Evaluation of Classification Methods for the Prediction of Hospital Length of

Stay using Medicare Claims data.

Length of stay in hospital determines the quality of care and safety of patients. It depends on various

factors and varies among medical cases with different conditions and complications. Length of stay

depends on factors such as age, sex, co-morbidities, time between surgery and mobilization, severity of

illness, etc [1]. It can be assumed that reduced length of stay hospital is associated with better health

results and good quality of care. Length of stay in hospital also depends upon the quick response to the

emergency medical case. The earlier the response the less is the length of stay and less likelihood of death
…show more content…
Unsupervised learning in data mining helps in clustering the data

by determining their similarity, helping patterns to emerge. The supervised learning is used to classify

new unknown data [4-7].

For the above scenarios, SPSS software [8] was for statistical analysis and WEKA 3.7 for classification

software [9].

The data that was used for analysis is a 2012 Medicare Provider Analysis and Review (MEDPAR) file.

The first scenario assumes that the patient has just been admitted and the information known about the

patient is as follows:

1>Patient demographics

2>Information about their hospital

3>Admission information

The following assumptions are made for the second scenario:

1>The patient has already been hospitalized for a few days.

2>Diagnosis is already known.

3>Procedures and other related hospital information.

Let’s consider the first LOS cut-off point is 4 days and the second LOS cut off point is 12 days. Here we

predict the LOS, if it is equal to 4days then it is approximately about 50% percentile of the LOS

distribution and LOS equal to 12 days is approximately 90% percentile of the LOS distribution. In this

experiment the performance of three classifiers are compared. The three classifiers used are Naïve Bayes,

AdaBoost and C4.5 decision tree. The first classifier used is Naïve Bayes it is a probabilistic classifier

based on Bayes theorem. The second classifier used is the Adaboost
Get Access