Multivariable prediction model for the need for surgery in horses with colic.

The research paper presents a multivariable prediction model developed from a survey of horse colic cases, aiming to predict the necessity of surgical treatment based on select variables. However, the model demonstrated insufficient fit to the validation data set.

Study Background and Objective

• The study, based on 1,965 horse colic cases from 10 equine referral centers in the United States, aimed to develop and validate a model that can predict the need for surgery in horse colic cases. The cases spanned from August 1985 to July 1986.

Methodology

• Two-thirds of the total cases (1,336) were randomly selected for model development, whereas the remaining one-third (629) were reserved for validating the model.
• Cases were classified as either ‘surgical’ – if a lesion was found during surgery or necropsy that required surgical resolution – or ‘medical’ – if no such lesion was found. The classification was used for subsequent model development and testing.
• An initial bivariable screening procedure was conducted, and variables that showed statistical significance (P<0.05) were incorporated into the model development.
• To handle missing values in the dataset, variables were only considered for model inclusion if fewer than 400 values were missing.
• A stepwise algorithm was used to construct a multivariable logistic regression model. A total of 640 cases were used and incorporated variables like rectal findings, signs of abdominal pain, peripheral pulse strength, and abdominal sounds.

Validation and Testing

• The resulting model’s likelihood ratio for surgery was calculated for each horse in the validation data set using the logistic regression equation.
• Using the Bayes theorem, the posttest probability was determined by using the calculated likelihood ratio as the test odds and the prevalence of surgical cases as an estimate of pretest odds.
• The model’s goodness-of-fit was also evaluated via a Hosmer-Lemeshow test, producing a significant chi-square value of 26.7 (P<0.001), indicating a poor fit of the model to the validation data set.

Conclusion

• Despite the comprehensive approach to creating a predictive model with significant variables, the model did not adequately meet the validation data set, as signified by the chi-square value from the Hosmer-Lemeshow test.