Development and Validation of a Predicting Model of All-Cause Mortality in Patients With Type 2 Diabetes Mellitus

  1. Vincenzo Trischitta, MD4,6
  1. 1Unit of Endocrinology, IRCCS Casa Sollievo della Sofferenza, San Giovanni Rotondo, Italy
  2. 2Unit of Biostatistics, IRCCS Casa Sollievo della Sofferenza, San Giovanni Rotondo, Italy
  3. 3Unit of Endocrinology, Department of Medical and Surgical Sciences, University of Foggia, Foggia, Italy
  4. 4Research Unit of Diabetes and Endocrine Diseases, IRCCS Casa Sollievo della Sofferenza, San Giovanni Rotondo, Italy
  5. 5Unit of Biostatistics, Consorzio Mario Negri Sud, Santa Maria Imbaro, Italy
  6. 6Department of Experimental Medicine, Sapienza University of Rome, Rome, Italy
  1. Corresponding author: Salvatore De Cosmo, s.decosmo{at}, or Vincenzo Trischitta, vincenzo.trischitta{at}
  1. S.D.C. and M.C. contributed equally to this work.


OBJECTIVE To develop and validate a parsimonious model for predicting short-term all-cause mortality in patients with type 2 diabetes mellitus (T2DM).

RESEARCH DESIGN AND METHODS Two cohorts of patients with T2DM were investigated. The Gargano Mortality Study (GMS, n = 679 patients) was the training set and the Foggia Mortality Study (FMS, n = 936 patients) represented the validation sample. GMS and FMS cohorts were prospectively followed-up for 7.40 ± 2.15 and 4.51 ± 1.69 years, respectively, and all-cause mortality was registered. A new forward variable selection within a multivariate Cox regression was implemented. Starting from the empty model, each step selected the predictor that, once included into the multivariate Cox model, yielded the maximum continuous net reclassification improvement (cNRI). The selection procedure stopped when no further statistically significant cNRI increase was detected.

RESULTS Nine variables (age, BMI, diastolic blood pressure, LDL cholesterol, triglycerides, HDL cholesterol, urine albumin-to-creatinine ratio, and antihypertensive and insulin therapy) were included in the final predictive model with a C statistic of 0.88 (95% CI 0.82–0.94) in the GMS and 0.82 (0.76–0.87) in the FMS. Finally, we used a recursive partition and amalgamation algorithm to identify patients at intermediate and high mortality risk (hazard ratio 7.0 and 24.4, respectively, as compared with those at low risk). A web-based risk calculator was also developed.

CONCLUSIONS We developed and validated a parsimonious all-cause mortality equation in T2DM, providing also a user-friendly web-based risk calculator. Our model may help prioritize the use of available resources for targeting aggressive preventive and treatment strategies in a subset of very high-risk individuals.

  • Received September 18, 2012.
  • Accepted February 21, 2013.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. See for details.