Risk factors, predictive models, and general work ability trajectory in patients with glioma

Abstract

Background: Glioma research has increasingly emphasized quality of life alongside traditional survival metrics, emphasizing functional outcomes, symptom burden, and social reintegration, including the ability to work. While previous studies focused on return-to-work rates, we assessed general work ability as a broader measure of work capacity. We aimed to develop predictive models for general work ability recovery, identify key risk factors, and explore long-term trajectories.

Methods: We conducted a retrospective cohort study of 342 patients with glioma (aged 18-64, WHO Grades 2-4) between March 2010 and December 2018. Work ability and symptoms were assessed using the M.D. Anderson Symptom Inventory-Brain Tumor Module (MDASI-BT), which was administered at months 1 and 3 postoperatively, then every three months up to 12 months, and at longer intervals thereafter. Logistic regression predicted 6-month general work ability recovery, and Cox models identified long-term risk factors. Long-term monitoring was conducted to evaluate the stability of work ability recovery across different WHO grades.

Results: 65.2% (223/342) regaining general work ability within 6 months post-surgery. Brain tumor-specific symptoms were stronger predictors of recovery than general symptoms. Predictive models achieved AUCs of 0.78 (pre-surgery) and 0.82 (post-surgery). Long-term monitoring showed recovery instability, with cumulative recovery rates for WHO Grades 2-4 at 82.1%, 50.8%, and 28.2%, respectively, while peaks at 50.8%, 28.3%, and 7.3%.

Conclusions: Brain tumor-specific symptoms significantly impact general work ability recovery. Recovery instability was observed across all patients, underscoring the importance of targeted symptom management, personalized care, and sustained follow-up to improve quality of life.

Keywords: glioma; risk factors; work.

Figure 1.

Flow diagram of patient selection. (A) The data were collected from the Department of Neurosurgery of Huashan Hospital, Fudan University, Shanghai, China. Patients aged 18–64 with WHO grade 2–4 were recruited in the basic database, as elder are defined as aged 65 and above according to WHO definition. Those with 6-month outcome data were incorporated into the Logistic dataset, and Logistic prediction models were established for both pre- and postoperative, and the follow-up data were used to construct Cox proportional hazards models. (B) Patients had first surgery in other hospitals were excluded for lack of follow-up data.

Figure 2.

Predictive models and the importance of symptom-related factors for general work ability recovery. Panels A and B show the ROC curves of preoperative and postoperative predictive models for 6-month general work ability recovery, including alternative models based on other combinations of variables. Panel C shows the SHAP summary plot, illustrating the association strength between selected variables and the likelihood of general work ability recovery. The width of the horizontal bars represents the magnitude of each variable’s impact on the association, the color gradient indicates the magnitude of variable values, and the x-axis orientation shows the association probability with higher (right) or lower (left) general work ability recovery. Abbreviations: SHAP = SHapley Additive exPlanations.

Figure 3.

Cumulative general work ability recovery analysis for postoperative patients with WHO Grades 2–4 gliomas. The figure shows the cumulative general work ability recovery rates over time for patients with glioma. The final cumulative rates of grades 2, 3, and 4 are 0.821, 0.508 and 0.282, respectively. The recovery rates decreased with increasing grade (P < .001), indicating a decreased likelihood of recovery in more advanced tumor grades.

Figure 4.

General work ability status trajectory post-surgery by WHO grade. The figure presents the trajectory of general work ability status over time for patients with WHO Grades 2–4 glioma in Panels A, B, and C, respectively. The stacked bar charts illustrate the distribution of patient statuses at different time points. The line graphs within each panel represent the low-interference proportion, calculated as Low-Interference Status / (Low-Interference Status + High-Interference Status + Deceased). The highest observed low-interference proportion was 50.8% for Grade 2 patients at 30 months, 28.3% for Grade 3 patients at 30 months, and 7.6% for Grade 4 patients at 12 months.