Development and Validation of Models to Predict Poor Health-Related Quality of Life Among Adult Survivors of Childhood Cancer

Abstract

Importance: Risk prediction models are important to identify survivors of childhood cancer who are at risk of experiencing poor health-related quality of life (HRQOL) as they age.

Objective: To develop and validate prediction models for a decline in HRQOL among adult survivors of childhood cancer.

Designs, setting, and participants: This prognostic study included 4755 adults from the Childhood Cancer Survivor Study (CCSS) diagnosed between January 5, 1970, and December 31, 1986, who completed baseline (time 0 [November 3, 1992, to August 28, 2003]) and 2 follow-up (time 1 [February 12, 2002, to May 21, 2005] and time 2 [January 6, 2014, to November 30, 2016]) surveys. Data were analyzed from June 19, 2019, to February 2, 2022.

Exposures: Sociodemographic, lifestyle, and emotional factors, and chronic health conditions (CHCs) were assessed at time 0 and time 1, and neurocognitive factors were assessed at time 1 to predict HRQOL at time 2 and a decline in HRQOL between time 1 and time 2. Impaired health states were defined as CHC grades 2 to 4 using the modified Common Terminology Criteria for Adverse Events, version 4.03, and mental and neurocognitive status as 1 SD or more below reference levels.

Main outcomes and measures: Health-related quality of life was operationalized using the Medical Outcomes Study 36-Item Short Form Health Survey Physical (PCS) and Mental (MCS) Component Summary and classified by optimal (≥40) or suboptimal (<40) at each point (main outcome). A decline in HRQOL was defined as a change from optimal to suboptimal between time 1 and time 2. Multivariable logistic regression identified factors associated with HRQOL decline. The cohort was randomly split into training (80%) and test (20%) data sets for model development and validation; the area under the receiver operating characteristic curve was used to evaluate prediction performance.

Results: A total of 4755 adults (mean [SD] age at time 0, 24.3 [7.6] years; 2623 [55.2%] women) were included in the analysis. Between time 1 and time 2, 285 of 3294 survivors (8.7%) had declining PCS and 278 of 3294 (8.4%) had declining MCS. Risk factors associated with PCS decline included female sex (odds ratio [OR], 1.67 [95% CI, 1.25-2.24]), family income less than $20 000 vs $80 000 or more (OR, 2.00 [95% CI, 1.21-3.30]), presence of CHCs (OR for neurological, 2.16 [95% CI, 1.51-3.10]; OR for endocrine, 2.25 [95% CI, 1.44-3.52]; OR for gastrointestinal tract, 1.89 [95% CI, 1.32-2.69]; OR for respiratory, 1.66 [95% CI, 1.06-2.59]; OR for cardiovascular, 1.53 [95% CI, 1.14-2.06]), and depression (OR, 1.79 [95% CI, 1.20-2.67]). Risk factors associated with MCS decline included unemployment vs full-time employment (OR, 1.68; [95% CI, 1.19-2.38]), current vs never cigarette smoking (OR, 2.03 [95% CI, 1.37-3.00]), depression (OR, 4.29 [95% CI, 2.44-7.55]), somatization (OR, 1.63 [95% CI, 1.05-2.53]), impaired task efficiency (OR, 1.90 [95% CI, 1.34-2.68]), and impaired organization (OR, 1.67 [95% CI, 1.12-2.48]). The areas under the receiver operating characteristic curve for the test models were 0.74 (95% CI, 0.67-0.81) for declining PCS and 0.68 (95% CI, 0.60-0.75) for declining MCS.

Conclusions and relevance: In this prognostic study of adult survivors of childhood cancer who experienced declining HRQOL, CHCs were associated with a decline in physical HRQOL, whereas current smoking and emotional and neurocognitive impairment were associated with a decline in mental HRQOL. These findings suggest that interventions targeting modifiable risk factors are needed to prevent poor HRQOL in this population.

Figure 1.. Predicted Probabilities by Predicted Risk Group of Suboptimal and Declining Health-Related Quality of Life

In each box and whisker plot, the top whisker indicates the maximum predicted probability; the bottom whisker indicates the minimum predicted probability. Blue lines indicate the mean of the predicted probability; orange lines, the median of the predicted probability; and orange dots, the observed probability. MCS indicates Mental Component Summary; PCS, Physical Component Summary.

Figure 2.. Observed Health-Related Quality of Life (HRQOL) Physical Component Summary (PCS) and Mental Component Summary (MCS) Scores and Their Changes by Predicted Risk Group

In each box and whisker plot, the top whisker indicates the maximum suboptimal HRQOL score (A and B) and maximum change of HRQOL score (C and D); the bottom whisker indicates the minimum suboptimal HRQOL score (A and B) and minimum change of HRQOL score (C and D). Blue lines indicate the mean of suboptimal HRQOL scores (A and B) and the change of HRQOL scores (C and D); orange lines indicate the median of suboptimal HRQOL scores (A and B) and the change of HRQOL scores (C and D).