Can the integrated and selected pattern approach effectively predict suicide rates? A study using internet search queries

Abstract

Background:

Suicide prevention has become a global public health and political issue. There have been active attempts to predict suicides (suicide rates and suicide numbers) using search queries from search engines.

Methods:

This study used spike-and-slab regression which is one of the sparse variable selection techniques, and we called this approach the ‘integrated and selected pattern approach.’ we used monthly national suicide data and relative search volume (RSV) of 51 search queries from Google Trends for the United States (US) and Japan from 2004 to 2019, and compared the accuracy of models using this approach against those using other approaches.

Results:

The model employing the integrated and selected pattern approach demonstrated the highest accuracy and stability of predicted values in predicting suicide rates and the number of suicides in the US. However, in the case of Japan, the accuracy and stability of the predicted values for this approach were lower than other approaches. Furthermore, even in the US, where the integrated and selected pattern approach outperforms other approaches at the country-level, it does not consistently outperform other models across all data subsets.

Limitations:

The integrated and selected pattern approach is not universally the most efficient method for predicting suicides across all linguistic, cultural contexts, and demographic groups. It remains crucial to emphasize the importance of comparing its efficacy with several other approaches in practical implementations.

Conclusion:

The integrated and selected pattern approach is one of the potentially effective methods for predicting suicides.