Novel Data Analytics Identify Predictors of Quality-of-Life Trajectories in Patients with AML or High-Risk Myeloid Neoplasms

Abstract

Background: Acute myeloid leukemia (AML) remains fatal in most patients (pts) with a 5-year survival probability of approximately 30% (less than 10% in pts aged 65 or older). Beyond survival, quality of life (QOL) can be significantly impaired by both disease and treatment-related factors. There is an urgent need to both characterize and identify factors predictive of QOL trajectories. Leveraging prospective data from 503 pts enrolled on an observational clinical trial, we implemented a novel statistical approach using non-supervised longitudinal clustering and ordinal logistic regression. We successfully identified: i) distinct QOL trajectories, ii) baseline factors independently associated with QOL trajectories. Methods: We analyzed data from pts with AML (90%) or high-risk myelodysplastic, myeloproliferative syndrome, or myelofibrosis (10%) enrolled on an observational clinical trial (NCT01929408) between 2013 and 2017 at 13 centers. QOL questionnaires were collected at the time of enrollment and approximately at months 1, 3, 6, 9, 12, 18, and 24 after study enrollment. Pts with a least six longitudinal data points available for QOL were included (n=503). Missing QOL data were imputed using linear interpolation. QOL was evaluated using the visual analog scale of the EQ5D (0, worst imaginable health state; 100, best imaginable health state). Deceased pts were considered to have a QOL of 0. Disease risk was assessed using the ELN 2017 criteria. Comorbidities and frailty were measured using the hematopoietic cell transplant comorbidity index (HCT-CI) and the NIH Toolbox 4-Meter Walk Gait Speed Test, respectively. QOL trajectories were clustered using non-supervised k-means based on Euclidian distances (TimeSeriesKMeans function of the tslearn library). Optimal cluster number was determined using the silhouette plot approach. Multivariable ordinal logistic regression was applied to predict QOL trajectories as a function of baseline variables using the rms R package. All analyses were done in Python (3.5) or R (4.0.2). Results: Six distinct QOL trajectories could be clustered and ordered from very good (n=109, 22%), good (n=76, 15%), intermediate 1 (n=49, 10%), intermediate 2 (n=86, 17%), poor (n=88, 17%), and very poor (n=95, 19%), as shown in the Figure. Comparisons of extreme clusters showed that pts in the very good QOL trajectory group were at baseline significantly younger (median age, 60 versus 66, respectively, p<0.001), had fewer comorbidities (median HCT-CI, 2 versus 4, respectively, p<0.001), less adverse AML cytogenetic/molecular risk (26% versus 48%, respectively, p=0.004), and were less frail (median walking speed, 0.89 versus 0.69m/s, p=0.009) compared to those in the very poor QOL trajectory group. A higher proportion of very good QOL pts received intensive chemotherapy (91% versus 62%, p<0.001) followed by consolidative allo-HCT (68% versus 11%, p<0.001) after achieving complete response/complete response with incomplete hematologic recovery (94% versus 40%, p<0.001). Multivariable modeling including baseline variables showed that older age (OR per 1-year increase, 1.34; 95%CI, 1.09-1.66; p=0.006), higher HCT-CI (OR per 1-unit increase, 1.66; 95%CI, 1.27-2.16; p<0.001), and frailty (OR per m/s increase, 0.8; 95%CI, 1.09-1.66; p=0.06) were associated with worse QOL trajectories. Favorable disease risk (favorable versus adverse: OR, 0.45; 95%CI, 0.28-0.71; p<0.001) and intensive therapy (OR, 0.54; 95%CI, 0.36-0.83; p=0.004) were independently associated with more favorable QOL trajectories. Conclusion: Unsupervised k-means clustering identified groups of pts with distinct QOL trajectories. Multivariable ordinal regression identified age, HCT-CI, disease risk and treatment intensity as independently associated with QOL trajectories. Favorable QOL trajectories were more likely to be achieved in younger, non-comorbid pts with favorable disease risk and receiving intensive therapy. Intensive therapy was independently associated with more favorable QOL trajectories even in older pts and in those with higher comorbidity burden. Probabilities of achieving the most favorable QOL trajectory in pts aged ≥60 after intensive therapy were <40% (<25% in case of adverse disease risk), indicating a critically unmet need to improve outcomes in older pts with high-risk myeloid neoplasms. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal