•We compared three TBCI and CCI in three different cohorts of older adults.•Excessive polypharmacy and high CCI predicted 1-year mortality better than TBCIs.•DDCI was also found to be a predictor of 1-year mortality, but not MCI and CDS.•Operating characteristics analysis showed poor accuracy for all the TBCI.•A more accurate TBCI to predict 1-year mortality should be developed. Comorbidity indexes were designed in order to measure how the disease burden of a patient is related to different clinical outcomes such as mortality, especially in older and intensively treated people. Charlson's Comorbidity Index (CCI) is the most widely used rating system, based on diagnoses, but when this information is not available therapy-based comorbidity indices (TBCI) are an alternative: among them, Drug Derived Complexity Index (DDCI), Medicines Comorbidity Index (MCI), and Chronic Disease Score (CDS) are available. This study assessed the predictive power for 1-year mortality of these comorbidity indices and polypharmacy. Survival analysis and Receiver Operating Characteristic (ROC) analysis were conducted on three Italian cohorts: 2,389 nursing home residents (Korian), 4,765 and 633 older adults admitted acutely to geriatric or internal medicine wards (REPOSI and ELICADHE). Cox's regression indicated that the highest levels of the CCI are associated with an increment of 1-year mortality risk as compared to null score for all the three samples. DDCI and excessive polypharmacy gave similar results but MCI and CDS were not always statistically significant. The predictive power with the ROC curve of each comorbidity index was poor and similar in all settings. On the whole, comorbidity indices did not perform well in our three settings, although the highest level of each index was associated with higher mortality.