Amsellem P, Bacon N, de Boer A, Groenwold RH, Hoes AW, Kirpensteijn J, Klungel OH, Kow K, Maritato K, Nielen M. Prev, Schmidt AF
Osteosarcoma (OS) is a malignant tumor of mesenchymal origin that produces osteoid. Given that the prognosis can vary considerably between dogs, we aimed to explore whether treatment could be tailored towards patient subgroups, characterized by their predicted risk of mortality. For the current study, a subset of five nonrandomized studies (400 subjects of whom 88 were dead at 5 months follow-up) was used from a previously published 20 study individual patient data meta-analysis. Missing data was dependent on observed variables and was imputed to correct for this dependency. Based on a previously published multivariable prognostic model, the 5-month mortality risk was predicted. Subsequently, in surgically treated dogs, using a logistic regression model with a random intercept for a study indicator, we explored whether chemotherapy effectiveness depended on predicted 5-month mortality risk. After adjustment for potential confounders the main effect of any chemotherapy was 0.48 (odds ratio) (95%CI 0.30; 0.78). Testing for chemotherapy by predicted 5-month mortality risk interaction revealed that the effects of any chemotherapy decreased with increasing predicted risk; interaction OR 3.41 (1.07; 10.84). Results from individually comparing carboplatin, cisplatin, doxorubicin and doxorubicin combination therapy to no chemotherapy, were similar in magnitude and direction. These results indicate that the main treatment effects of chemotherapy do not necessarily apply to all patients.