Deep Learning-based Time-to-event Analysis of Depression and Asthma using the All of Us Research Program

Abstract

While there is a growing recognition of the association between depression and asthma, few studies have leveraged deep learning-based (DL-based) models in a retrospective cohort study with a large sample size. We analyzed the association between depression and asthma among 239,161 participants of the All of Us Research Program through DL-based, logistic regression, and Cox Proportional Hazards (CoxPH) models. We used SHAP values to help interpret DL-based models and c-index to evaluate model performance. Results suggest a significant odds ratio for depression in asthma. The c-indices for the CoxPH, DeepSurv, and DeepHit models were 0.619, 0.625, and 0.596, respectively. SHAP indicated a different set of important variables when compared with the CoxPH model. In conclusion, we provide strong evidence of a positive relationship between depression and asthma. Also, DL-based models did not outperform the CoxPH model on the c-index. Sex at birth and income may play important roles in occurrence of depression in asthma patients.

Figure 1.

Forest Plot of Odds Ratios from the Multivariate Logistics Model for Participants with Depression. This cross-sectional multivariate logistic regression suggests asthma has the strongest association with the diagnosis of depression, with an OR of 3.59 (95%CI: [3.50, 3.68]) after adjustment. (Reference levels: Race White, Age < 30, Ethnicity Non-Hispanic, Annual Income <50K, Sex female, Never Smoked.)

Figure 2.

Kaplan-Meier Curve plot for the diagnosis of depression. With a median time to depression of 21.39 years, the KM curve suggests a faster rate of depression in the first ten years compared with later time. The longest follow-up for non-asthmatics was about 6 years.

Figure 3.

Forest Plot of Hazard Ratios from CoxPH model: The CoxPH model suggests BMI and smoking could be risk factors for depression among asthma participants, whereas Asian race and male sex at birth might be protective. (Reference levels: Race White, Age <30, Ethnicity Non-Hispanic, Annual Income <50K, Sex female, Never Smoked.)

Figure 4.

DL-based model architecture for DeepSurv and DeepHit. With three hidden layers for 64 neurons each, DeepSurv is able to predict hazards at each time point. With ten time intervals and three hidden layers for 64 neurons each, DeepHit predicts hazards at each time interval.

Figure 5.

Global SHAP Values and a random participant SHAP value in DeepSurv and DeepHit. Both DL-based models agreed that income and sex at birth played important roles in the diagnosis of depression among participants with asthma. In the same individual plot of DeepSurv (Figure 5C) and DeepHit (Figure 5D), lower income, ever-smokers, and the female sex contribute to a higher hazard output, and lower BMI is a protective factor.

Figure 6.

Bump Chart for ranking in different models. The ranking of features changes through CoxPH, DeepSurv, and DeepHit models. Generally, the three models agree well with each other. The Asian race got a high rank in the CoxPH model but a lower rank in DL-based models. All three models agreed that income and sex can impact the risk of depression.