Trends in epidemiology of lung cancer. A single center study

baseline, they had standardized Comprehensive Geriatric Assessment, physical examination and blood tests. They were followed-up for death during one year. Deaths were identified through medical charts or Public Records Office. Scores were built for each patient and assessed using calibration (Hosmer–Lemeshow test) and discrimination (C-statistic). 95% confidence intervals were estimated using a bootstrap model with 1000 resampling. Results: We reviewed 13,794 eligible articles to identify 11 generic one-year mortality scores published between 1968 and 2011. None was developed in cancer population setting and only 4 were independently validated. We were able to build 6 scores: the Charlson Comorbidity Index (CCI, 17 items) and Romano's implementation (Romano's Charlson, 15 items), the G-CIRS (14 items), the scores of Walter et al. (6 items), Elixhauser et al. (21 items) and Gagne et al. (20 items). All but the G-CIRS had a category for cancer (metastatic or not). Of 645 patients included, 67 were lost to follow up and 272 had at least one missing data, leaving a 306 patients study population. Mean age was 79.8 (SD, 5.6) and 164 (53.6%) were women. Of 277 (90.5%) solid cancer patients, 76 (27.4%) had colorectal cancer, 72 (26.0%) breast cancer and 31 (11.2%) prostate cancer. Overall, 134 (48.4%) had metastatic cancer. All but the CCI score (p Hosmer–Lemeshow=0.18) had adequate calibration. Although the 5 remaining scores had C-statistic of 0.70 or greater (Romano's Charlson: 0.75; 95% CI: 0.66–0.84, G-CIRS: 0.76; 0.67– 0.85, Elixhauser et al.: 0.79; 0.70–0.87), only two had 0.80 or greater (Gagne et al.: 0.80; 0.72–0.87, Walter et al.: 0.83; 0.73–0.91). Conclusion: We demonstrated a very good accuracy of two scores i.e. Walter et al. [1] and Gagne et al. [2], in a French elderly cancer patient population. Due to a few numbers of items (6 versus 20), the score of Walter et al. may be the best choice for clinical practice.