TY - JOUR T1 - Using syndrome mining with the Health and Retirement Study to identify the deadliest and least deadly frailty syndromes JF - Scientific Reports Y1 - 2020 A1 - Yi-Sheng Chao A1 - Chao-Jung Wu A1 - Hsing-Chien Wu A1 - Hui-Ting Hsu A1 - Tsao, Lien-Cheng A1 - Cheng, Yen-Po A1 - Lai, Yi-Chun A1 - Wei-Chih Chen KW - Epidemiology KW - Geriatrics AB - Syndromes are defined with signs or symptoms that occur together and represent conditions. We use a data-driven approach to identify the deadliest and most death-averse frailty syndromes based on frailty symptoms. A list of 72 frailty symptoms was retrieved based on three frailty indices. We used data from the Health and Retirement Study (HRS), a longitudinal study following Americans aged 50 years and over. Principal component (PC)-based syndromes were derived based on a principal component analysis of the symptoms. Equal-weight 4-item syndromes were the sum of any four symptoms. Discrete-time survival analysis was conducted to compare the predictive power of derived syndromes on mortality. Deadly syndromes were those that significantly predicted mortality with positive regression coefficients and death-averse ones with negative coefficients. There were 2,797 of 5,041 PC-based and 964,774 of 971,635 equal-weight 4-item syndromes significantly associated with mortality. The input symptoms with the largest regression coefficients could be summed with three other input variables with small regression coefficients to constitute the leading deadliest and the most death-averse 4-item equal-weight syndromes. In addition to chance alone, input symptoms’ variances and the regression coefficients or p values regarding mortality prediction are associated with the identification of significant syndromes. VL - 10 SN - 2045-2322 JO - Scientific Reports ER -