113 Background: Health-Related Quality of Life (HRQoL) is an important issue for elderly patients with colon cancer. We created the expert system which allows to predict low level of HRQoL and accessed it’s quality by using several simulation studies. Methods: We performed a systematic review to figure out the known factors associated with low level of HRQoL in elderly colon cancer patients. The searches were performed in PubMed. We accessed the possible impact of several factors affecting HRQoL, including symptoms, comorbidities and treatment toxicity. All relevant factors were included in prediction model. We assigned the different weights to different factors based on evaluation of clinical studies to develop the logistic regression and Markov stochastic model later. As we needed a binary dependent variable we performed the ROC analysis to figure out an optimal cutoff of HRQoL. Then we simulated a partly virtual dataset based on elderly colon cancer patients diagnosed in Davidovskiy Hospital to evaluate the prediction model quality. All statistical calculations were performed in RStudio. The simulation part was performed using simFrame R package. Results: Twenty two studies with a total number of 2516 patients were included in our systematic review. The 39 factors with different weights were included prediction model with different weights assigned. The weights range varied from 1 to 18.6. The adjusted proportion of summary score's variance (R2 ) varied from 0.09 to 0.47 in univariate analysis. The final logistic regression model quality was moderate: the Nagelkerke R-square coefficient was 57.9. However, the developed model showed a 76% sensitivity and 61% specificity in predicting of lower HRQoL level. Conclusions: Our prediction model allows to prospectively manage of elderly colon cancer patients, making the emphasis on HRQoL. However, the present study has some restrictions: simulation nature of internal validation, possible underestimating of the rare events impact. The long-term comprehensive approach with external validation using large real data analysis is needed to evaluate our prediction model.
Read full abstract