. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A few clinical features improve the prediction of mortality and cardiovascular outcomes in patients with type 2 diabetes Maria Masulli 1 , Giuseppe Lucisano 2 , Enzo Bonora 3 , Stefano Del Prato 4 , Angela A. Rivellese 1 , Stefano Signorini 5 , Paolo Mocarelli 5 , Gabriele Riccardi 1 , Olga Vaccaro 1 *, and Antonio Nicolucci 2 ; on behalf of the TOSCA.IT Investigators 1 Department of Clinical Medicine and Surgery, Federico II University, Via S. Pansini 5, 80131 Naples, Italy; 2 CORESEARCH – Center for Outcomes Research and Clinical Epidemiology, via Tiziano Vecellio 2, 65124 Pescara, Italy; 3 Division of Endocrinology, Diabetes and Metabolism, University and Hospital Trust of Verona, Piazzale Aristide Stefani 1, 37129 Verona, Italy; 4 Department of Clinical & Experimental Medicine, University of Pisa, Lungarno Pacinotti 43, 56126 Pisa, Italy; and 5 University Department of Laboratory Medicine, Hospital of Desio, via Giuseppe Mazzini 1, 20832 Desio (MB), Italy Received 5 May 2020; revised 18 June 2020; accepted 14 July 2020; online publish-ahead-of-print 27 September 2020 Type 2 diabetes (T2DM) is a disease with heterogenous phenotypic manifestations. 1 Distinct subgroups, each with salient metabolic char- acteristics, risk of complications, and response to treatments, have been identified. 2,3 An accurate phenotyping could improve outcome prediction and prevention strategies. However, which phenotyping parameters may ensure the greatest yield in terms of the cost- effectiveness of interventions remains unclear. 4,5 Many risk scores have been developed for people with T2DM, but their performance is modest in most populations; innovative statistical methods may provide further insight. 5,6 The study aims are to establish, in a cohort with T2DM, the feasibility and clinical relevance of a risk stratification based on variables habitually assessed in clinical practice, and identify significant modifiable risk predictors that may guide group-specific interventions by using the RECursive Partitioning and Amalgamation (RECPAM) analysis. We studied 2820 patients with complete data-set, enrolled in the TOSCA.IT trial (NCT00700856) designed to evaluate the cardiovas- cular (CV) effects of Pioglitazone or Sulfonylureas (SUs) as add-on to metformin. 7 The study primary endpoint was a composite of all- cause death, non-fatal myocardial infarction, non-fatal stroke, and ur- gent coronary revascularization. This was a pragmatic trial, and the selection of all-cause death as part of the primary outcome is of rele- vance because this is the most robustly ascertained endpoint, and ar- guably one with the greatest clinical significance. All outcomes were adjudicated in blind. Median follow-up was 57.3 months (IQR 42.2– 60.2). Age and diabetes duration were 62.3 ± 6.5 and 8.5 ± 5.7 years; glycated haemoglobin (HbA1c) was 7.7 ± 0.5 %; average systolic blood pressure and LDL cholesterol were 132 ± 15 mmHg and 103 ± 32 mg/dl, 11% reported a prior CV event, 18% were smokers; all patients were on oral antidiabetic medications, 69% were taking antihypertensive drugs, 57% lipid-lowering medications. To identify subgroups of patients with different risk of outcome, we used the RECPAM method. 8 This tree-based method integrates the advan- tages of the main effects of standard Cox regression and tree- growing techniques. At each partitioning step, the method chooses the covariate and its best binary split to maximize the difference in the risk of the outcome. The algorithm stops when user-defined con- ditions are met (stopping rules). We considered a minimum set of 30 events and 200 subjects per node. The variables entered into the model were age (considered as a global adjustment variable), gender, smoking, diabetes duration, dyslipidaemia, hypertension, retinopathy, previous cardiovascular disease (CVD), anti-platelets drugs, family history of CVD, body mass index (BMI), HbA1c, systolic blood pres- sure, low-density lipoprotein (LDL) cholesterol, non-HDL (high- density lipoprotein) cholesterol, urinary albumin-to-creatinine ratio (ACR), C-reactive protein. Five subgroups with increasing risk of out- come were identified (Figure 1). Women (class 1) showed the lowest risk and were taken as reference. Among men, ACR > 22 mg/g and HbA1c >7.7% (class 5) conferred the highest risk (HR 4.03; 95% CI 2.55–6.39). A significant excess risk was also documented in men with ACR>22 mg/g and HbA1c <_7.7% (class 4; HR 3.26; 95% CI 2.07–5.15). Among men with ACR<_ 22 mg/g, non-HDL cholesterol >125 mg/dL (class 3) was associated with significantly higher risk of outcome (HR 2.61; 95% CI 1.76–3.88), whereas men in class 2 (ACR<_22 mg/g and non-HDL cholesterol <_125 mg/dL) showed a risk not significantly different from that of women (HR 1.43; 95% CI 0.91–2.26). The Cox-regression analysis with the RECPAM classes forced in confirmed the increasing risk associated with RECPAM classes and highlighted an independent role of age, BMI, smoking, and previous CV events. The patients’ characteristics in the five risk classes are given in Table 1. Patients in classes 4 and 5 had higher ACR, BMI, waist circumference, and blood pressure values, and the * Corresponding author. Tel: þ 39 -081 7463665, Fax: þ39 081 5453136, Email: ovaccaro@unina.it Published on behalf of the European Society of Cardiology. All rights reserved. V C The Author(s) 2020. For permissions, please email: journals.permissions@oup.com. European Journal of Preventive Cardiology (2021) 28, e1–e3 RESEARCH LETTER doi:10.1093/eurjpc/zwaa002 Risk Prediction/Assessment & Stratification Downloaded from https://academic.oup.com/eurjpc/article/28/18/e1/5912215 by guest on 21 December 2022