The prognostic value of patient-reported outcomes in allogeneic hematopoietic stem cell transplantation: exploratory analysis of a randomized nutrition intervention trial

The original RCT and the current study were approved by the Regional Committee for Medical and Health Research Ethics South East Norway (#S-09136c 2009/2115), and the Data Protection Supervisor, Oslo University Hospital and conducted in accordance with the Declaration of Helsinki. All patients gave written informed consent. The original RCT is registered in ClinicalTrials.​gov, ID NCT01181076.

One hundred and seventeen patients (intervention n=57, control n=60) were included in this RCT conducted at the Department of Haematology, Oslo University Hospital from August 2010 to February 2017. The aims of the RCT were to assess the impact of optimized energy and protein intake compared to routine hospital nutrition support on global QoL and clinical outcomes three months after allo-HSCT. A detailed description of the RCT and main clinical outcomes have been reported previously and showed no significant differences among the two study groups on any of the QoL-C30 scales or items [17].

In short, the nutritional intervention started upon hospital admission with optimization of energy and protein intake until discharge (usually after 3–5 weeks) [16]. The patients in the intervention group had their daily energy and protein requirement estimated according to World Health Organization recommendations, i.e., 126 to 167 kJ (30 to 40 kcal) per kg each day and 1.5–2.0 g protein/kg/each day [18] and validated by measuring the patients’ energy expenditure with indirect calorimetry, adding an activity factor. Oral intake was monitored by the patient’s self-reports with additional enteral parenteral nutrition if the estimated intake was insufficient, i.e,. lower than the estimated energy needs. Patients in the control group received a standard amount of parenteral nutrition combined with oral intake if possible.

At enrollment, all the participants provided information on the following sociodemographic factors: age, sex, education, and marital status. The electronic health records were used to provide information on disease-, transplant- and treatment-specific information such as diagnosis, conditioning regime, admission dates, duration of hospitalization, days alive, and duration of out of the hospital within the first year after allo-HSCT.

The European Organisation for Research and Treatment of Cancer Core Quality of Life Questionnaire (QLQ-C30) was used to assess QoL at baseline (i.e., on day 8 or 7 before allo-HSCT) [19]. The QLQ-C30 includes five functional scales (role, physical, cognitive, emotional and social), three symptom scales (nausea/vomiting, fatigue, and pain), and six single items (insomnia, dyspnea, appetite loss, constipation, diarrhea, and financial problems). This questionnaire also includes a global health status/quality of life (QoL) scale. All scales and single items are transformed into standardized scores ranging from 0 to 100, with higher scores for the functioning scales and global health status/QoL scale indicating better outcomes, while higher scores on the symptom scales and single items indicate greater symptom severity [19].

Clinical- and transplant-related data were registered daily during hospitalization and later retrieved from the medical health records. This included diagnosis and progression of disease, conditioning regime, donor information, stem cell source, the HCT-CI, and EBMT scores. The HCT-CI score summarizes the presence of 17 comorbidities, and as such, does not denote the patient perspective [20]. Patients are classified in three risk groups based on the sum score (low risk = 0 points, intermediate risk =1-2 points, or high risk > 3 or more points), which correlates highly with 2-year NRM [21]. The EBMT risk score includes five factors: stage of the disease, age of the patient, time from diagnosis, donor type, and donor-recipient sex combination. This risk score complements the HCT-CI classification by emphasizing transplant-related factors [22]. In addition, the Eastern Cooperative Oncology Group (ECOG) score for assessment of physical performance was used at baseline. This score ranges from 4 (fully active patient with no performance limitations) to 0 (bedridden, completely disabled) [23].

The patients were assessed at baseline, at 3 and 6 weeks and then at 3, 6, 9, and 12 months after transplantation. In these exploratory analyses, we use clinical data and PROs at baseline.

All descriptive statistics were performed and reported overall and by study group affiliation (intervention and control). Due to the exploratory nature of the current study, we assessed all the scales from the QLQ-C30 questionnaire without selecting any primary scales. For each scale, we used Cox proportional hazard models to investigate possible associations of the corresponding pre-treatment score with the risk of dying (for any cause) over 1 year, and logistic regression for 1-year NRM, respectively. For each clinical outcome, we first ran a univariable model using sociodemographic and clinical variables. Then, we ran a multivariable model, including the statistically significant variables in univariable analysis, i.e., age, living arrangements (living alone vs. not), stem cell source (peripheral blood vs. bone marrow), and EBMT score. In addition, we forced key variables that we deemed important from a clinical point of view into the multivariable model, i.e., gender, study group (intervention vs. control), HCT-CI score, body mass index and donor type (related vs. unrelated). Then, for each outcome, we ran the same multivariable model also including the pre-treatment score of each QLQ-C30 scale. The additional information provided by a scale was evaluated by the likelihood ratio test, testing the null hypothesis that the scale did not significantly increase the model fit when added to sociodemographic and clinical variables. Values are reported as hazard ratio (HR) or odds ratio (OR). The significance level was set at 0.05 with no adjustment for multiple testing, and all statistical tests were two-sided. All analyses were performed by SAS software v.4 (SAS Institute Inc., Cary, NC, USA).

In the original RCT, 173 patients ≥ 18 years of age undergoing allo-HSCT with myeloablative conditioning were assessed for eligibility. Of these, 119 patients consented and were randomly assigned to receive the nutrition intervention or the standard total parental nutrition. Two patients in the intervention group were excluded from further analysis (Fig. 1), leaving 117 patients for the intention-to-treat analysis (intervention n=57, control n=60).

Most patients (74%) were categorized as low risk according to the HCT-CI score, and acute myeloid leukemia was the predominant diagnosis Table 1. The different baseline characteristics were evenly distributed between the two study groups, indicating that the randomization was adequate. Thirty-five (30%) patients died during the first year, 15 patients of these due to relapse. There was no significant difference in 1-year survival between the two study groups. For descriptive purposes, the QoL baseline scores for all patients combined and per groups are shown in Table 2.

Univariable analyses of 1-year OS are shown in Supplementary Tables 1 and 2 whereas the corresponding multivariable analyses are given in Table 3. Both analyses are exploratory, and hence, p-values are not adjusted for multiplicity meaning that statistical significance only suggests an association. The multivariable analysis identified only the HCT-CI score (HR=1.414; 95% CI, 1.137 to 1.759, p=0.002) and the EBMT score (HR=1.387; CI, 1.020 to 1.887, p=0.037) as statistically significant, with the HCT-CI score also being significant in the univariable analysis. Global QoL was significant in the univariable analysis (HR=0.975; 95% CI, 0.953 to 0.998, p=0.04), (Supplementary Table 2), but this was not the case in the multivariable analysis.

Univariable analysis on the prognostic significance of the clinical-sociodemographic variables are shown in Supplementary Table 3. The statistically significant clinical-sociodemographic variables related to 1-year NRM included age (OR=1.043; 95% CI, 1.002 to 1.085, p=0.038), living alone (OR=4.210; 95% CI, 1.384 to 12.809, p=0.011), stem cell source (OR=3.918; 95% CI, 1.221 to 12.567, p=0.022), and EBMT score (OR=2.082; 95% CI, 1.364 to 3.178, p=0.001). Supplementary Table 4 shows the univariable analysis of the predictive significance of PROs. Global QoL (OR=0.975; 95% CI, 0.953 to 0.998, p=0.034) was the only statistically significant predictor of 1-year NRM.

The exploratory multivariable model without PROs contained nine variables, of which four were identified as candidates for being prognostic for 1-year NRM: living alone (OR=7.980; 95% CI, 1.701 to 37.446, p=0.009) HCT-CI (OR=1.616; 95% CI, 1.092 to 2.392, p=0.016), EBMT risk score (OR=2.891; 95% CI, 1.493 to 5.599, p=0.002), and stem cell source (OR=4.609; 95% CI, 1.024 to 20.740, p=0.046) (Table 4). P-values were not adjusted for multiplicity. After having run this model by adding separately each scale of the QLQ-C30, only appetite loss (i.e., higher appetite loss) was statistically associated with a slightly lower chance of 1-year NRM (OR=0.95; 95% CI, 0.908 to 0.994, p=0.026) (Table 5). The global QoL scale did no longer remain significant here, as it was in the univariable analysis (data not shown).