The new comorbidity index: A development and validation study

Student thesis: Master thesis (including HD thesis)

  • Sinna Pilgaard Ulrichsen
  • Rikke Beck Nielsen
4. term, Mathematics, Master (Master Programme)
When the prognosis of a disease is studied it has been shown that comorbid diseases have an influence on the outcome. To adjust for this influence a comorbidity index can be used. The most used index is the Charlson comorbidity index, which was developed in 1987 on a small cohort of patients from the medical service at New York Hospital. The aim of this thesis is to investigate the ability of the Charlson comorbidity index to predict mortality on a cohort of pneumonia patients, and to develop and validate a new comorbidity index.

We used a cohort of hospitalized pneumonia patients from the Danish National Registry of Patients. We validated the Charlson comorbidity index by including it in a logistic regression with 30 day mortality as an outcome and assessing the performance.

Both logistic regression, naive Bayes and classifications trees were used to develop new indexes. When using the logistic regression method we updated the weights on the original Charlson diseases, included three new diseases, included first degree interaction terms and a variable for 'time since diagnosis'.

The naive Bayes method and classification trees were used as alternatives to the logistic regression model. Indexes made by these methods included the original Charlson diseases and the three new diseases.

We validated the indexes by assessing their performance for both 30 day and 1 year mortality. Their crude performance was assessed by the Pearson chi squared test of a contingency table for the index and mortality. To assess their adjusted performance we included each index in a logistic regression adjusted for sex and age.

Our analysis showed that the Charlson comorbidity index predicted death among pneumonia patients well, and therefore it is still usable.

All of our developed indexes performed well and most of them better than CCI. Our analysis showed that four of our indexes were better than the rest. For these indexes their complexity increased with performance. Choosing the best index out of these is therefore a balance between performance and simplicity and depends on the situation at hand.
Publication date1 Jun 2010
Number of pages205
Publishing institutionDepartment of Mathematical Sciences, Aalborg University
ID: 31973054