Citation

Parker J, Cuthbertson C, Loveridge S, Skidmore M, Dyar W. J. Affect. Disord. 2016; 213: 9-15.

Affiliation

Agricultural, Food, and Resource Economics, Michigan State University, 458 W Circle Dr., Suite 908, Cook Hall, East Lansing, MI 48824-1039, USA. Electronic address: dyarwill@msu.edu.

Copyright

(Copyright © 2016, Elsevier Publishing)

DOI

10.1016/j.jad.2016.10.038

PMID

28171770

Abstract

BACKGROUND: Vital statistics on the number of, alcohol-induced death (AICD) drug-induced death (DICD), and suicides at the local-level are only available after a substantial lag of up to two years after the events occur. We (1) investigate how well Google Trends search data explain variation in state-level rates in the US, and (2) use this method to forecast these rates of death for 2015 as official data are not yet available.

METHODS: We tested the degree to which Google Trends data on 27 terms can be fit to CDC data using L1-regularization on AICD, DICD, and suicide. Using Google Trends data, we forecast 2015 AICD, DICD, and suicide rates.

RESULTS: L1-regularization fit the pre-2015 data much better than the alternative model using state-level unemployment and income variables. Google Trends data account for substantial variation in growth of state-level rates of death: 30.9% for AICD, 23.9% for DICD, and 21.8% for suicide rates. Every state except Hawaii is forecasted to increase in all three of these rates in 2015. LIMITATIONS: The model predicts state, not local or individual behavior, and is dependent on continued availability of Google Trends data.

CONCLUSIONS: The method predicts state-level AICD, DICD, and suicide rates better than the alternative model. The study findings suggest that this methodology can be developed into a public health surveillance system for behavioral health-related causes of death. State-level predictions could be used to inform state interventions aimed at reducing AICD, DICD, and suicide.

Copyright © 2017. Published by Elsevier B.V.

Language: en

Keywords

Behavioral health; Forecasting; Google Trends; Regional analysis; Substance abuse; Suicide