A score prediction model for predicting the heterogeneity symptom trajectories among lung cancer patients during perioperative period: a longitudinal observational study

Abstract

Introduction: Patients undergoing video-assisted thoracoscopic surgery (VATs) for lung cancer (LC) frequently experience prolonged symptoms that can significantly affect their quality of life (QoL).

Patients and methods: This study employed a longitudinal observational design. The MDASI and QLQ-C30 were utilized to evaluate symptoms and QoL one day before surgery, as well as at 1 day, 2 weeks, and 1, 2, and 3 months post-surgery. Latent class growth modeling (LCGM) was employed to identify heterogeneous trajectories. By Logistic regression analysis, a score prediction model was developed based on predictive factors, which was internally validated utilizing 1000 bootstrap samples. The SHaply Additive Explanations (SHAP) was used to calculating the contribution of each factor.

Results: 205 participants participated in this study. The predominant postoperative complaints included fatigue, shortness of breath, pain, and coughing. Two distinct classes of symptom trajectories were identified: 'severe group' and 'mild group'. Four independent predictors of heterogeneous symptom trajectories were used to develop a scoring model. The area under the receiver operating characteristic curve for this model was 0.742 (95% CI: 0.651-0.832). And the calibration curves demonstrated strong concordance between anticipated probability and actual data (mean absolute error: 0.033). Furthermore, the decision curve analysis (DCA) indicated higher net benefit than other four single factors. SHAP highlighted WBC and surgical duration time as the most influential features.

Conclusions: We established a score model to predict the occurrence of severe symptom trajectories 3 months postoperatively, promoting recovery by advancing rehabilitation plan based on preoperative and surgical situation.

Registration: ClinicalTrials.gov (ChiCTR2100044776).

Keywords: Lung cancer; prediction model; quality of life; surgery; symptom burden; symptom trajectory.

Figure 1.

The total composite symptom scores of the two groups.

Figure 2.

(A) Proportions of patients exhibiting better and unimproved composite symptom scores on the MDASI-TCM. (B) Physical function scores on the QLQ-C30 from enrollment to 3 months. (C) Role function scores in the QLQ-C30 from enrollment to 3 months. (D) The scores of emotional function from enrollment to 3 months. (E) The scores of social function in QLQ-C30 from enrollment to 3 months. (F) Cognitive function scores in the QLQ-C30 from enrollment to 3 months. QLQ-C30, European Organization for the Research and Treatment of Cancer Quality of Life Questionnaire Core 30. MDASI-TCM, MD Anderson Symptom Inventory-Traditional Chinese Medicine; **p < 0.05, ***p < 0.001.

Figure 3.

(A) ROC curve of the model (B). Model calibration curves. The x- and y-axes depict the expected and observed occurrence of heterogeneous symptom trajectories, respectively. (C). DCA curves of the model. The black line indicates that no cases were in the severe group, while the gray line indicates the opposite. (D). Calibration curves for 1000 bootstrap internal validations of the model. DCA, decision curve analysis. AUC, area under the receiver-operating characteristic curve; ROC, receiver-operating characteristic.

Figure 4.

(A) Attributes of characteristics in SHAP. Each line represents a feature, and the abscissa is the SHAP value. (B). Feature importance ranking as indicated by SHAP. The matrix diagram describes the importance of each covariate in the development of the final prediction model. SHAP, SHaply Additive Explanations.