Scientific research Results - Study HRQoL trajectories stage IV cancer

Patients with stage IV cancer often experience multiple, long-term, and sequential symptoms. Experiencing symptoms simultaneously has a detrimental impact on the patients’ health-related quality of life (HRQoL), even at relatively low levels of symptoms. HRQoL can vary over time and differ between patients. Therefore, the aim of this study was twofold: 1) to identify clusters of patients with stage IV cancer with distinct trajectories of HRQoL, and 2) identify characteristics of patients at high risk of developing impaired HRQoL. 

Methods

We pooled data from two longitudinal studies assessing HRQoL in patients with stage IV cancer: the eQuipe and the SYMPRO-Lung studies. In both studies, the European Organisation for Research and Treatment of Cancer (EORTC) quality of life questionnaire core 30 (QLQ-C30) was used to assess HRQoL. Of the EORTC QLQ-C30, we used the five functioning scales and six symptom scales to assess distinct trajectories of HRQoL. To identify clusters of patients with distinct trajectories of HRQoL development we used group-based multi-trajectory modelling (GBMTM). We evaluated the model fit using the following criteria: Bayesian Information Criterion (BIC), average posterior probabilities of group assignment (APPA, at least 0.7), group sizes (at least 5%), odds of correct classification (at least 5) and clinical relevance of the added group or differences between groups. 
Multivariable logistic regression analyses were performed to identify the patients at high risk of impaired HRQoL. For this, we dichotomized the clusters resulting from the GBMTM models into a normal functioning and an impaired functioning group and a normal symptom burden and a high symptom burden group, respectively. Next, we assessed which (combination of) demographic and clinical characteristics were predictive of membership of the impaired HRQoL clusters using a backward selection procedure. Validation of the model was performed using calibration (calibration plot) and discrimination techniques (Harrell’s C-statistic). For this, we randomly split the pooled data into a training (70%) and validation cohort (30%). We imputed missing data using Multivariate Imputation by Chained Equations (MICE).

Results

The included 841 patients had a mean age of 65.5 (SD 9.8) years and 50% was male. The main areas of tumour location were the thorax (47%) or gastrointestinal organs (18%). The majority of patients had one or two comorbidities (56%) and the major treatments included chemotherapy (55%) and immune therapy (35%). 

 

Distinct trajectories of HRQoL

We identified a six-group model with cubic order for the functioning scales (figure 1) and a five-group model with cubic order for the symptom scales (figure 2).  
 

Figure 1: Distinct trajectories of functioning

Open in new window

 

Figure 2: Distinct trajectories of symptoms

Open in new window

We dichotomized the clusters resulting from the GBMTM models. For the functioning scales, the clusters ‘impaired and deteriorating functioning’, ‘impaired but stable functioning’ and ‘stable moderate functioning’ were grouped into an impaired functioning group (41%), while the other three clusters were grouped into a normal functioning group (59%). For the symptom scales, the clusters ‘severe symptom burden’ and ‘moderate symptom burden’ were grouped into a high symptom burden group (38%) and the other three clusters into a normal symptom burden group (62%). The patients belonging to both the impaired functioning group and the high symptom burden group were classified as patients with impaired HRQoL. This group consisted of 247 patients, which is 29% of the study sample.

 

Characteristics of patients at high risk of impaired HRQoL

We split the study cohort into a training cohort (70%) and a test cohort (30%). In the training cohort, we identified the characteristics of patients with an impaired HRQoL using a multivariable regression model. Table 1 shows the final model after a backward selection procedure. 

 

Table 1: Final multivariable logistic regression model in training cohort. Abbreviations: CI, confidence interval; HRQoL, health-related quality of life; OR, odds ratio

 

The final regression model was validated in the training cohort and the test cohort. The discriminatory ability was moderate (C-statistic 0.691 for training cohort and 0.685 for the test cohort). The calibration plots are shown in figure 3.

 

Figure 3: Calibration plots in the training cohort (A) and test cohort (B)