Vision-related quality of life compared to generic measures in retinoblastoma survivors

Retinoblastoma (RB) is an intraocular (eye) cancer that forms in the developing retina in very young children and can affect one or both eyes [1]. The yearly worldwide incidence is one in every 16,000 live births [1] with an improved survival rate of 97% over the past decade in the United States (U.S.) [2]. Treatment depends on the size, location, and laterality of the tumor(s), which can be unilateral or bilateral, resulting from spontaneous or hereditary gene mutation [3]. The number one treatment goal in the management of RB is to save the life of the child; secondary treatment goals include vision preservation and ocular salvage. Nonetheless, survivors can be left monocular (unilateral), visually impaired, or blind. RB affects physical, mental, and social health aspects of life. Hence, studies examining health-related quality of life (HRQOL) in RB survivors are needed but to date have been limited [4‐6] with only one previous study using a vision-related quality of life (VRQOL) measure [7].

Assessing HRQOL is important during RB treatment due to decisions negotiated between the provider and the parent. Many parents are hesitant to accept enucleation (surgical removal of the eye), a common treatment for RB, due to fear of social stigma and cultural factors, instead opting for alternate, possibly less-effective therapies [8]. This can result in tumor metastases and not infrequently eventual enucleation after failed attempts at globe salvage. These decisions are even more difficult when both eyes are affected. Knowledge of HRQOL in RB survivors can be utilized by providers in discussions with parents during early treatment decision making.

Results of HRQOL studies of RB survivors have found varied results that may reflect the use of a variety of measures [4]. Generic HRQOL measures are useful to compare disease groups with healthy counterparts. However, sensitivity to specific HRQOL effects of the disease may be lacking. There is no disease-targeted HRQOL measure that has been psychometrically evaluated for use in the RB population [5]. Limitations in the methodologic quality of pediatric ophthalmology and oncology measures and applicability of content to the RB population have been identified [5]. However, it is uncertain how well a generic HRQOL measure captures the impact of RB compared to a VRQOL measure.

In this study, we compare VRQOL in adolescent and young adult (AYA) unilateral versus bilateral RB survivors. We hypothesize that VRQOL will be worse for those with bilateral than unilateral RB due to more extensive treatment and visual impairment with bilateral disease. In addition, we assess associations among the VRQOL, generic HRQOL, and overall QOL.

Demographic and clinical characteristics and VRQOL scores by disease laterality were summarized using mean/median, standard deviation, and range for continuous variables and frequencies and percentages for categorical variables. Differences in baseline characteristics between unilateral and bilateral RB survivors were examined using chi-square tests for categorical variables and Mann–Whitney U for continuous variables. Internal consistency reliability (Cronbach’s alpha coefficient) [27] was estimated for the NIH Toolbox® VRQOL and PROMIS®-29 domains. IRT pattern-based T scores for the VRQOL were estimated using the Health Measures Automated Scoring Service and PROMIS® HRQOL scales were estimated using the PROMIS® Assessment Center Scoring Service.^SM Normality was assessed with Kolmogorov–Smirnov and Shapiro–Wilk tests. Most measurement domains and clinical variables were not normally distributed. Mann–Whitney U tests were used to compare differences in VRQOL between unilateral and bilateral RB survivors.

A correlation of 0.10–0.29 was defined as small, 0.30–0.49 as medium, and > = 0.50 as large per Cohen’s recommendations [28, 29]. We hypothesize medium to large correlations between NIH Toolbox® VRQOL and PROMIS® domains: (1) psychosocial compared to anxiety, depression, and the mental health summary score; (2) role performance with ability to participate in social roles and activities, physical function, and physical health summary score; and (3) distance and near vision with physical function, ability to participate in social roles and activities, and physical health summary scores, respectively. Product-moment and Spearman rank order correlation coefficients are reported.

We examined correlations of the PROMIS® overall QOL item with the PROMIS®-29 and NIH Toolbox® VRQOL scale scores. Next, we regressed the overall QOL item on the PROMIS-29 and Toolbox measures to obtain explained variance. We estimated four models with different subsets of variables: (1) PROMIS-29 domains; (2) VRQOL domains; (3) PROMIS-29 and VRQOL domains; and (4) PROMIS-29 summary scores. We employed five-fold leave-one-out cross-validation with the CVPRESS statistic [30] to identify the subset of variables to include in each model. We report ordinary least squares regression model results for the selected independent variables.

Detailed description of participants including flow diagram of recruitment has been previously published [31]. A total of 184 RB survivors were identified from a clinic database, 61 were unable to be contacted and 123 were screened for eligibility. Of the 123 RB survivors, 7 were screened ineligible, 13 declined to participate, and 2 did not complete the study questionnaires for a total of 101 RB survivors enrolled. We do not have information about the 13 individuals who declined to participate and the 2 who did not complete study requirements, but the study included a racially and ethnically diverse sample with a majority of participants identifying as Hispanic/Latino.

Among 101 RB survivors, 57% had unilateral disease and 43% had bilateral disease (Table 1). The median age of survivors was 17 with a range of 14–26 years, 50% were female, 65% were Latino/Hispanic, and 27% with middle class socioeconomic status. The median age at diagnosis was 15 months (range 1–108), 88% had an enucleation in at least one eye, and 13% had a previous family history of RB. Out of those who received genetic testing (n = 86), 40 (46%) had a germline pathogenic variant in the RB1 gene. The majority had normal vision defined as > 20/40 in the better seeing eye (86%); 8% were visually impaired and 6% were blind defined as 20/400–20/40 and < 20/400, respectively. There were no statistically significant differences between unilateral and bilateral groups regarding gender, race/ethnicity, and enucleation laterality. Bilateral survivors were older (median 18 versus 16 years) at the time of study but younger (median 9 months versus 20.5) at diagnosis. All unilateral survivors had normal visual acuity while as expected, bilateral survivors showed more family history (23%) of RB and the presence of a pathogenic germline RB1 gene mutation (74%). Given the small sample sizes for these subgroups any significant difference is non-trivial. For example, the difference in age between the unilateral and bilateral subgroups is a medium effect (d = 0.67).

Coefficient alphas for the NIH Toolbox® scales in this study were as follows: 0.88 color vision, 0.95 distance vision, 0.93 near vision, 0.84 ocular symptoms, 0.85 psychosocial, and 0.83 role performance. Coefficient alpha for the PROMIS®-29 Profile scales in this study ranged from 0.69 (physical function) to 0.96 (pain interference).

The VRQOL domain scores in unilateral and bilateral RB survivors are listed in Table 2. Unilateral RB survivors had significantly better VRQOL scores for color vision, distance vision, near vision (all p < 0.001), and role performance (p = 0.009) than bilateral survivors. Differences in ocular symptoms and psychosocial domains were not significant between the groups. Bilateral survivors scored worse than the NIH Toolbox developmental sample of 819 adults (T score mean = 50) on all domains except ocular symptoms while unilateral survivors scored below the sample mean only on psychosocial well-being and role performance.

Correlations among the VRQOL, PROMIS®-29, and the overall QOL measures are shown in Table 3. All VRQOL scales were significantly correlated with each other except color vision and ocular symptoms. Most scales had medium to large correlations with each other except color vision which had weak correlations with the ocular symptoms and psychosocial scales. The largest correlation (r = 0.838) was between near and distance vision scales. There were large correlations of the PROMIS®-29 physical function scale with the VRQOL color vision (r = 0.623), distance vision (r = 0.532), and role performance (r = 0.570) scales. Role performance also had a large correlation with the PROMIS®-29 ability to participate in social roles (r = 0.509). Distance vision (r = 0.563), near vision (r = 0.527), and role performance (r = 0.578) were correlated largely with the PROMIS®-29 physical health summary score. There were medium-sized correlations between the VRQOL psychosocial scale and all PROMIS®-29 scales except sleep disturbance (r’s = 0.328-0.488). The overall QOL item had a medium correlation with psychosocial (r = 0.391) and role performance scales (r = 0.359) and a small correlation with distance vision (r = 0.197).

Thirteen of 15 correlations of the overall quality of life item with the PROMIS-29 and NIH Toolbox VRQOL scales were statistically significant: depression (r = −0.595, p < 0.0001), mental health summary score (r = −0.575, p < 0.0001), anxiety (r = −0.465, p < 0.0001), ability to participate in social roles (r = 0.444, p < 0.0001), fatigue (r = −0.435, p < 0.0001), psychosocial (r = 0.407, p < 0.0001), sleep disturbance (r = −0.386, p < 0.0001), role performance (r = 0.357, p = 0.0003), physical health summary (r = 0.321, p = 0.0011), pain interference (r = -0.312, p = 0.0015), ocular symptoms (r = 0.273, p = 0.0057), physical function (r = 0.232, p = 0.0110), and distance vision (r = 0.214, p = 0.0057).

Table 4 provides a summary of the multivariate analysis. The PROMIS®-29 scales explained 38% (adjusted R-square) of the variance in overall QOL, with depression (ß = −0.495) and ability to participate in social roles and activities (ß = 0.216) having significant unique associations with overall QOL. The mental health summary score was significantly associated with overall QOL (ß = 0.575) in the second regression model, explaining 32% of the variance. Among the VRQOL domains, only distance vision, psychosocial, and role performance domains significantly correlated with overall QOL. The psychosocial domain (ß = 0.407) was the only VRQOL domain significantly uniquely associated with overall QOL, explaining 16% of the variance. The final model explained 39% of the variance with depression (ß = −0.375) and the mental health summary score (ß = 0.303) having significantly unique associations with overall QOL.

When controlling for demographic (age and SES) and clinical variables (medical conditions, enucleation, visual acuity), there was no change to the PROMIS®-29 model. In the second model, SES was significantly associated with overall QOL along with the mental health summary score, explaining 36% of the variance. SES was also significant in the third (VRQOL) model along with the psychosocial domain, explaining 22% of the variance. With SES included, the final model explained 42% of the variance with depression, mental health summary score, and the VRQOL psychosocial domain having unique associations with overall QOL.

Four of the 6 NIH Toolbox® VRQOL scales correlated significantly with the single overall QOL item. The magnitude of the associations were smaller than those for the PROMIS-29 scales. This is consistent with the fact that QOL measures often reflect mental rather than physical health [26]. The color vision domain was not significantly associated with overall quality of life. It may not be as applicable in this population due to the sparing of the optic nerve in treatment, although this is not always possible [38]. Thus, color vision deficits in this sample may be due to overall visual impairment or blindness. Distance vision was significantly related to overall quality of life, but near vision was not. This could be because the distance vision scale includes questions related to driving and driving is a key factor in adolescents’ and young adults’ independence. Nonetheless, distance vision was only weakly correlated with overall QOL.

The regression models accounted for no more than 39% of the variance of overall QOL with 61% unexplained or not captured in the RB population. Many of the predictive domains were psychosocial, depression, or mental health related. Our findings could reflect data collection during a global pandemic and civil unrest affecting mental health more pronouncedly in the AYA age group. The results also reflect the conceptual similarity and potential overlap of overall QOL with mental health [39]. The VRQOL appears to not add any additional value to the PROMIS®-29 in predicting overall QOL. While the use of a vision-targeted measure such as the NIH Toolbox® VRQOL does add information in the assessment of HRQOL in AYA RB survivors by incorporating vision-related issues that can impact their daily life, it does not replace the value of generic measures, such as the PROMIS®-29 instrument.

This study adds new information about VRQOL in AYA RB survivors, but it has limitations. Our cross-sectional design precludes the ability to make any causal inferences and limits the interpretation of associations. The lack of a control group for comparison with other types of vision-related disorders limits our ability to make unbiased estimates of effect. The generalizability of our findings may be limited due to our ethnically diverse sample from the Los Angeles community, with most other samples being predominantly Caucasian. Furthermore, our sample reflected the treatment era of a decade ago which may be reflective of the previous use of radiation; additionally, with improvements in treatment modalities there are fewer enucleations required now compared to previous decades. Non-response and selection bias cannot be ruled out as only survivors seeking active follow-up and those who responded to letters and phone calls were recruited. In addition, data were self-reported and with some visually impaired participants requiring assistance in completion of the measures, subjects may have chosen socially acceptable answers rather than being truthful. Moreover, the relatively small sample size limited our options for cross-validation of the regression models. Furthermore, since data collection occurred during the COVID-19 pandemic, psychosocial effects on the AYA population may have played a role in the predominance of mental health factors influencing overall QOL. But perceptions of QOL are largely reflective of mental health so it is not surprising that the PROMIS®-29 and NIH Toolbox VRQOL mental health scales were most strongly associated with overall QOL [39].

Bilateral RB survivors report worse VRQOL than unilateral survivors, but no differences were found in specific ocular symptoms or psychosocial well-being between groups. Medium to large correlations were found between a targeted measure (NIH Toolbox® VRQOL) and a generic (PROMIS®-29 Profile) HRQOL measures. Our findings indicate that the NIH Toolbox® VRQOL adds to the assessment of HRQOL in AYA RB survivors. Although the NIH Toolbox® VRQOL does not significantly add to the prediction of overall QOL beyond the PROMIS®-29 Profile, it provides information about specific vision-related effects on daily life of AYA RB survivors.