Long-term effectiveness and cost-effectiveness of high versus low-to-moderate intensity resistance and endurance exercise interventions among cancer survivors

Purpose This study aimed to evaluate the long-term effectiveness and cost-effectiveness of high intensity (HI) versus low-to-moderate intensity (LMI) exercise on physical fitness, fatigue, and health-related quality of life (HRQoL) in cancer survivors. Methods Two hundred seventy-seven cancer survivors participated in the Resistance and Endurance exercise After ChemoTherapy (REACT) study and were randomized to 12 weeks of HI (n = 139) or LMI exercise (n = 138) that had similar exercise types, durations, and frequencies, but different intensities. Measurements were performed at baseline (4–6 weeks after primary treatment), and 12 (i.e., short term) and 64 (i.e., longer term) weeks later. Outcomes included cardiorespiratory fitness, muscle strength, self-reported fatigue, HRQoL, quality-adjusted life years (QALYs) and societal costs. Linear mixed models were conducted to study (a) differences in effects between HI and LMI exercise at longer term, (b) within-group changes from short term to longer term, and (c) the cost-effectiveness from a societal perspective. Results At longer term, intervention effects on role (β = 5.9, 95% CI = 0.5; 11.3) and social functioning (β = 5.7, 95%CI = 1.7; 9.6) were larger for HI compared to those for LMI exercise. No significant between-group differences were found for physical fitness and fatigue. Intervention-induced improvements in cardiorespiratory fitness and HRQoL were maintained between weeks 12 and 64, but not for fatigue. From a societal perspective, the probability that HI was cost-effective compared to LMI exercise was 0.91 at 20,000€/QALY and 0.95 at 52,000€/QALY gained, mostly due to significant lower healthcare costs in HI exrcise. Conclusions At longer term, we found larger intervention effects on role and social functioning for HI than for LMI exercise. Furthermore, HI exercise was cost-effective with regard to QALYs compared to LMI exercise. Trial registration This study is registered at the Netherlands Trial Register [NTR2153 [http://www.trialregister.nl/trialreg/admin/rctview.asp?TC=2153]] on the 5th of January 2010. Implications for Cancer Survivors Exercise is recommended to be part of standard cancer care, and HI may be preferred over LMI exercise.


Introduction
Supervised exercise can contribute to counteracting the negative side effects of cancer and its treatments [1]. Systematic reviews demonstrated safety and beneficial effects of exercise on physical fitness [2], fatigue [3], and health-related quality of life (HRQoL) [4], during and after cancer treatment. However, previous studies predominantly reported shortterm effects [4,5]. The few studies that included a longerterm follow-up (≥ 6 months) showed that the benefits of exercise were maintained for HRQoL [4], but not for fatigue [5]. For other outcomes (e.g., physical fitness), longer-term effects are unclear. Therefore, more research on the longer-term effectiveness of exercise in cancer survivors is warranted.
As resources are scarce, decisions on the implementation of healthcare programs are guided not only by their health effects but also by their additional costs in relation to these effects (i.e., cost-effectiveness). Therefore, it is important that state-of-theart cost-effectiveness analyses of healthcare programs are performed [6]. Cost-effectiveness analyses of exercise interventions in cancer survivors are scarce [7,8]. A systematic review compared exercise interventions to usual care in patients with various diseases, including cancer, found acceptable incremental cost-effectiveness ratios or cost savings [7]. Similar results were found by a systematic review evaluating the costeffectiveness of multidimensional cancer rehabilitation programs [8]. However, despite the fair methodological quality of the reviewed studies, the heterogeneity across interventions hampered solid conclusions about their cost-effectiveness.
The present study reports the effectiveness and costeffectiveness of the randomized controlled Resistance and Endurance exercise After ChemoTherapy (REACT) study at longer term (i.e., 64 weeks) [9]. At short term (i.e., 12 weeks), high intensity (HI) and low-to-moderate intensity (LMI) exercise interventions significantly improved cardiovascular fitness and HRQoL and reduced fatigue compared to a wait list control (WLC) group, with some indication for a dose-response relationship for exercise intensity on cardiorespiratory fitness [9]. Also, HI and LMI exercise were equally beneficial in counteracting fatigue [9]. This study aimed to evaluate the longer-term effectiveness and cost-effectiveness of HI versus LMI exercise for physical fitness, fatigue, and HRQoL.

Setting and participants
Detailed methods, including sample size calculations, of the REACT study have been reported previously [9,10]. Briefly, REACT is a multicenter randomized controlled trial (RCT) in cancer patients recruited from nine Dutch hospitals between 2011 and 2013. The Medical Ethics Committee of the VU University Medical Centre approved the study. Patients aged ≥ 18 years with histologically confirmed breast, colon, ovarian, cervix or testis cancer, or lymphomas with no indication of recurrent or progressive disease who had completed (neo-)adjuvant chemotherapy with curative intent were eligible and invited to participate. Exclusion criteria were the following: being unable to perform daily activities; presence of cognitive disorders, severe emotional instability, and diseases that hamper patients' capacity of carrying out HI exercise; and being unable to read and write Dutch. Written informed consent was obtained from all participants prior to participation.

Randomization
Following baseline assessments, participants were stratified by cancer type and hospital and randomly assigned to HI exercise, LMI exercise, or WLC using random numbers tables [9]. Shortly after randomization, HI and LMI participants commenced their 12-week exercise program. WLC participants were also randomly allocated to HI or LMI exercise, but started exercising after the 12-week follow-up assessment. Allocation sequence was concealed from the clinical and research staff. Due to the interventions' nature, participants and physiotherapists were not blinded.

Exercise interventions
HI and LMI exercise interventions had similar exercise types, durations, and frequencies, but differed in intensity (Table 1). Exercise sessions were given twice per week during 12 weeks and supervised by a trained physiotherapist. Both exercise programs included six resistance exercises targeting large muscle groups (i.e., vertical row, leg press, bench press, pull over, abdominal crunch, and lunge), with a training volume of two sets of ten repetitions [10]. Workload per exercise was defined by an indirect one-repetition maximum (1-RM) measurement. Following a warm-up, the physiotherapist estimated a workload at which the patient was expected to perform four to eight repetitions, taking into consideration age, gender, and height [10]. Furthermore, both programs included two types of endurance interval exercises. During weeks 1-4, patients cycled 2 × 8 min with alternating workloads (defined by the maximum short exercise capacity (MSEC) estimated by the steep ramp test [11]). Patients in the HI exercise group alternated 30 s at 65% of MSEC with 60 s at 30% of MSEC. From the fifth week onwards, the 30 s at 65% of MSEC was alternated with 30 s at 30% of MSEC. The workload for the LMI exercise group alternated between 45 and 30% of MSEC in a similar way. During weeks 5-12, one additional endurance interval session was added substituting one 8-min interval of cycling. This interval session consisted of 3 × 5 min of aerobic exercise at constant workload (defined by the heart rate reserve using the Karvonen formula) [12]. The use of the Karvonen formula allowed patients to perform these aerobic exercises using different ergometers (e.g., cycle ergometer, treadmill). Physiotherapists applied behavioral motivational counseling techniques to overcome possible exercise barriers and to encourage participants to obtain and maintain a physically active lifestyle. At 4, 10, and 18 weeks after intervention completion, three booster sessions (i.e., supervised workout sessions) were provided to motivate participants to maintain their exercise engagements.

Measurements
Socio-demographic data were collected by self-report. Clinical information was obtained from medical records. Physiotherapists documented session attendance in exercise logs. Outcomes were assessed at baseline, and after 12 and 64 weeks, except for dual energy X-ray absorptiometry (DXA), which was only performed at baseline and 64 weeks. Detailed descriptions of the assessments and their measurement properties are provided elsewhere [10,13]. Physical tests were performed by an independent assessor.

Primary outcomes
Cardiorespiratory fitness was measured during a maximal cyclometer exercise test aiming to achieve peak oxygen uptake 1-RM, one-repetition maximum; MSEC, maximum short exercise capacity; HRR, heart rate reserve; a Every 4 weeks (week 1, 5 and 9), the physiotherapist evaluated training progress and adjusted the workload accordingly b Exercises included vertical row, leg press, bench press, pull over, abdominal crunch, and lunge) c Exercises were accompanied with BORG scores and heart rate monitors to guide the physiotherapists. In the occasion that the training intensity seemed too high or too low, the 1-RM, MSEC, or HRR was reassessed d In the first four weeks, 30 s at 65% of MSEC was alternated with 60 s at 30% for HI, and from the fifth week onwards, intensity was alternated every 30 s. The workload for the LMI exercise group was alternated between 45 and 30% of MSEC in a similar way (peakVO 2 , in mL/kg/min) within 8-12 min [14] following a ramp protocol, in which breath-by-breath gas exchange was measured continuously. After each test, peakVO 2 (i.e., highest oxygen consumption values averaged over a 15-s interval within the last 60 s), peak power output (peakW, in watt), and the ventilatory threshold (determined by the oxygen equivalent method [14]) were recorded. Hand-grip strength was assessed using a JAMAR hand-grip dynamometer [15] and the mean score (in kg) of three attempts with the participants' dominant hand was used for further analyses. Lower body function was assessed using the 30-s chair-stand test [16]. The total number of times participants raised to a full stand in 30 s was reported. Fatigue was assessed using the Multidimensional Fatigue Inventory (MFI) [17], including five subscales: general fatigue, physical fatigue, reduced physical activity, reduced motivation, and mental fatigue.

Secondary outcomes
HRQoL was measured using the European Organisation Research and Treatment of Cancer-Quality of Life questionnaire-Core 30 (EORTC-QLQ-C30) [18] and anxiety and depression by the Hospital Anxiety and Depression Scale (HADS) [19]. Physical activity (PA) was objectively assessed by accelerometers (Actigraph) using vertical accelerations converted into counts/minute. Body mass index (BMI) was calculated from measured body height and weight. Body composition was determined using percentage of total body fat mass (%FM), lean mass (%LM), and lumbar spine (L1-L4) bone mineral density (BMD), measured by DXA with a Hologic Discovery DXA scanner. Quality-adjusted life years (QALYs) were estimated using the EQ-5D-3L [20]. EQ-5D-3L health states were convertedintoutilities usingtheDutchtariff [21].QALYs werecalculated using linear interpolation between measurement points.

Cost measures
Intervention costs were micro-costed [22,23]. Attendance of exercise and booster sessions were registered, intervention providers' time investments were valued using their gross hourly salaries (including overhead), and material costs were estimated using invoices. All other cost categories were assessed using 3monthly questionnaires, with 3-month recall periods. Healthcare costs included costs due to primary and secondary healthcare use, and medication. Dutch standard costs were used to value healthcare use [23]. Medication use was valued using unit prices of the Royal Dutch Society of Pharmacy [24]. Informal care (i.e., care by family/friends) was valued using a shadow price [23]. Absenteeism was assessed using participants' reports of their number of absence days and, in case of partial absence, their percentage of normal working hours worked. Using the friction cost approach (FCA), absenteeism costs were valued with ageand gender-specific price weights [23,25]. The FCA assumes that costs are limited to the friction period (i.e., period needed to replace a sick-listed worker = 23 weeks) [23,25]. Unpaid productivity (e.g., volunteer work) losses were valued using the aforementioned shadow price [23]. Sports costs included expenses on memberships and equipment. All costs were converted to 2012 euros (€) [26].

Statistical analyses
Differences in outcomes between HI and LMI exercise interventions at longer-term follow-up were assessed using linear mixed model analyses with a two-level structure (i.e., participants were clustered within hospitals). Both interventions were simultaneously regressed on the longer-term value of the outcome, adjusted for the baseline value, age, gender, and timing of intervention (i.e., direct start or WLC). To check whether missing data affected the results, sensitivity analyses were conducted on an imputed dataset for peakVO 2 , hand-grip strength, and fatigue (SA1). Missing data were multiple imputed using predictive mean matching, stratified by group allocation [27]. The imputation model was specified according to White et al. [27]. Twenty different datasets were created. Pooled estimates were calculated using Rubin's rules [27].
To evaluate within-group changes in HI and LMI exercise interventions from short-term to longer-term follow-up, we conducted linear mixed models for repeated measurements (i.e., repeated measurements were clustered within patients, which were clustered within hospitals). This model simultaneously regressed the intervention effect on short term and longer term and included time and the interaction between time and exercise group as determinants and age, gender, and the outcome's baseline value as covariates.
Cost-effectiveness analyses were performed from the societal perspective using the multiple imputed datasets [27]. Betweengroup differences were estimated for total and disaggregated costs. Total cost and effect differences were estimated using linear mixed model analyses, adjusted for baseline, age, gender, and intervention timing. Incremental cost-effectiveness ratios (ICERs) were calculated by dividing the adjusted total cost differences by those in effects [6]. Uncertainty around cost differences and ICERs was estimated using bias-corrected (BC) bootstrap intervals (5000 replications, stratified by hospital) [28]. Costeffectiveness planes [29] and cost-effectiveness acceptability (CEA) curves were constructed [30]. A post hoc analysis was performed applying a healthcare perspective and a sensitivity analysis (SA2) was conducted assuming that all scheduled exercise sessions needed tobe paid for, rather than only those attended.
As disease recurrence-which may influenced quality of life and healthcare costs-occurred more often during followup in LMI exercise, additional sensitivity analyses (SA3-4) were performed. We excluded patients with disease recurrence (n = 17) in order to check whether disease recurrence affected the results of the main effectiveness (i.e., between-group difference-SA3) and cost-effectiveness analyses (SA4). All primary analyses were performed according to intention-to-treat. Costs and effects beyond 1 year were discounted at a rate of 4 and 1.5%, respectively [23]. Effectiveness analyses were performed in SPSS (v22.0) and multiple imputation and cost-effectiveness analyses in STATA (v12.0). P < 0.05 was considered significant.
Data availability At present, raw data of the REACT study forms part of a PhD project, including current data on long-term (cost-)effectiveness. As a consequence, currently, we are unable to publish this dataset. However, the REACT dataset is included in the internationally shared POLARIS database [31], and researchers who are interested to collaborate are invited to prepare a paper proposal.

Screened (n=793)
Patients not eligible (n=38; 5%) (cognitive disorders or severe emotional instability n=7; serious diseases that hampers patients' capacity of carrying out HI exercise n=21; inability to understand the Dutch language n=7; already participating in an exercise study n=1; complications due to cancer treatments n=1) At longer term, intervention effects on role functioning (β between-group difference = 5.9, 95% CI = 0.5; 11.3) and social functioning (β between-group difference = 5.7, 95% CI = 1.7; 9.6) were larger for HI than for LMI exercise (Table 3). No other significant between-group differences were found at longer term. Results of the sensitivity analyses (SA3) were comparable (data not shown).
No significant within-group changes were found for peakVO 2 and HRQoL between short and longer terms for both HI and LMI exercise interventions, indicating that the intervention-induced improvements at short term were maintained at longer term (Table 3). For HI exercise, role functioning (β within-group change = 5.5, 95% CI = 0.3; 10.6), hand-grip strength (β within-group change = 1.4, 95% CI = 0.6; 2.2), and BMI (β within-group change = 0.3, 95% CI = 0.03; 0.5) increased from short to longer term, and lower body muscle function increased both in HI and LMI exercise interventions (HIβ within-group change = 1.4, 95% CI = 0.7; 2.2, LMI-β within-group change = 1.2, 95% CI = 0.5; 1.9). For both groups, significant within-group changes were found for fatigue and anxiety, such that they returned to baseline levels. No significant within-group changes from short to longer term were found for depression and objectively measured PA.
Total societal costs did not differ significantly between HI and LMI exercise interventions (β = − 2429€, 95% CI = − 5798; 933). In HI exercise, healthcare costs were significantly lower (β = − 2056€, 95% CI = − 3816; − 443) and intervention costs were significantly higher (β = 40€, 95% CI = 8; 75) than in LMI exercise (Table 4). For QALYs, an ICER of − 87,831 was found, indicating that HI exercise was associated with a cost saving of 87,831€/QALY gained, compared with LMI exercise (Table 5). When societal decision-makers are not willing to pay anything per unit of effect gained, the probability of HI exercise being cost-effective compared with LMI exercise was 0.87. This probability increased to 0.91 at a willingness-to-pay of 20,000€/QALY and reaching 0.95 at 52,000€/QALY. For hand-grip strength, the probability of cost-effectiveness increased as the willingness-to-pay increased, from 0.87 to 0.95 at 58.000€/kg, while it decreased for peakVO 2 and general fatigue (data not shown). From a healthcare perspective, results were more favorable for HI exercise as shown by higher probabilities of cost-effectiveness, e.g., if healthcare decision-makers are not willing to pay anything per unit of effect gained, the probability of costeffectiveness was 0.97 for all outcome measures. When we assumed that all scheduled exercise sessions needed to be paid for SA2, we found comparable results. When patients who had a disease recurrence during follow-up were excluded from the analyses (SA4), the mean difference in total societal costs between HI and LMI exercise interventions was smaller (i.e., − 1366€ versus − 2429€). Additionally, HI exercise had slightly lower probabilities of being cost-effective in comparison with LMI exercise (i.e., 0.89 versus 0.96 at 80,000€/QALY). However, the societal cost difference was in favor of HI exercise, in both the main analysis and SA4, and the differences in effect were comparable.

Discussion
At longer term (i.e., 64 weeks), effects on role and social functioning were significantly larger for HI than for LMI exercise. Within-group changes showed that interventioninduced improvements in cardiorespiratory fitness and HRQoL found at short term were successfully maintained at longer term for HI and LMI exercise interventions, whereas fatigue returned to baseline levels. Also, HI exercise was costeffective for QALYs, compared to LMI exercise. The mean improvements in peakVO 2 after exercise (HI 4.3 mL/kg/min, LMI 3.5 mL/kg/min) at longer term were in line with the mean improvement of 3.3 mL/kg/min found in cancer survivors after supervised exercise, as reported in a previous meta-analysis [2]. In contrast with the tendency of a dose-response relationship of exercise intensity at short term [9], no significant differences in peakVO 2 were found at longer term between HI and LMI exercise. Nevertheless, the exercise-induced benefits on peakVO 2 at short term were successfully maintained over time in both exercise groups. Although this is hopeful, we should acknowledge that, compared to healthy adults, the patients' level of peakVO 2 at longer term was still Bpoor^ [14]. Apparently, a 12-week exercise program is too short for patients to fully recover to normative values. On the other hand, there may be a possibility that patients do not fully recover after having received multiple cancer treatments.
At longer term, hand-grip strength and lower body muscle function were not significantly different between HI and LMI exercise interventions. A previous meta-regression analysis revealed that the effects of resistance training on muscle strength may be more dependent on volume than on intensity [32]. Additional head-to-head comparisons of exercise programs with different exercise parameters (i.e., frequency, intensity, type, time) are therefore warranted to define the optimal exercise dose on muscle strength for cancer survivors. Furthermore, increases in both strength outcomes between short and longer terms for both groups suggest that these improvements result from increased uptake of daily activities during follow-up.
At longer term, self-reported fatigue did not differ significantly between HI and LMI exercise interventions, and in both groups, it returned to baseline values between short and longer terms. This lack of sustainable improvements in fatigue is in line with previous studies [3] and may be related to the patients' low self-efficacy in managing fatigue, particularly while resuming daily activities without supervision and support from a physiotherapist [33]. On the other hand, self-reported fatigue in a longitudinal study is also susceptible to Bresponse-shift bias,^resulting from a change in the internal standard of fatigue perception throughout the cancer continuum [34].
We found a significant better social and role functioning for HI exercise compared to LMI exercise at longer term. In addition, longer-term effects on global QoL and physical functioning tended to be larger for HI than for LMI exercise, but this was not significant. Overall, current findings reveal a possible dose-response relationship of exercise intensity for some HRQoL domains among cancer survivors. Hence, a previous meta-analysis reported significant exercise effects on global QoL and social functioning, but not on role functioning [4].   Based on our significant effects on role functioning, it may be hypothesized that participants gain confidence from completing a HI exercise program [35], resulting in improvements in a person's role in society. Furthermore, the exercise-induced benefits on HRQoL were successfully maintained over time in both interventions, despite the return to baseline levels of fatigue. This indicates that besides fatigue, which is found to mediate the exercise effect on HRQoL [36], other factors also contribute to HRQoL. The lack of a significant difference between HI and LMI exercise interventions in psychological distress at longer term is in contrast with a previous meta-analysis reporting small but significant reductions in depression and anxiety after exercise at short and longer terms, compared to usual care [4,37]. Yet, our study lacked a non-exercise group and the mean values for both outcomes were already low at baseline, leaving little room for improvement. Furthermore, from short to longer term, anxiety returned to baseline in both groups, despite the beneficial short-term intervention effects on anxiety after HI exercise [9]. So, HI exercise might be more effective in reducing anxiety compared to LMI exercise; however, sustainability is lacking which may reflect the vulnerability of psychosocial recovery [38].
Comparable with our short-term findings [9], there were no significant differences between HI and LMI exercise interventions in body composition and objectively measured PA at the longer term. This may be related to the design of our exercise interventions. To successfully reduce fat mass, complementary dietary changes may be required [39] and improving and maintaining PA may require specific behavioral change techniques (e.g., motivational interviewing [40], goal setting [41]). Our finding that BMI significantly increased from short to longer term in HI exercise is unexpected, and its clinical meaningfulness may be questioned, as it was not supported by changes in %FM and %LM.
At the lower bounds of the Dutch and UK willingness-topay threshold (i.e., 20,000 and 24,400€/QALY gained, respectively), the probability of HI exercise being cost-effective compared to that of LMI exercise was ≥ 0.91 and increased even more with increasing willingness-to-pay values. Thus, at longer term, HI exercise can be considered cost-effective compared with LMI exercise for QALYs, if decision-makers are willing to accept a probability of cost-effectiveness of 0.91 and to pay 20,000€/QALY. The relatively high probabilities of cost-effectiveness seemed to be related to lower healthcare costs in HI exercise. Although smaller, the healthcare costs were still lower after excluding patients with disease recurrence, and HI exercise remained cost-effective. Current results support previous results of a systematic review showing acceptable cost-effectiveness ratios for cancer rehabilitation programs that produced significant health gains [8] compared to usual care. As willingness-to-pay thresholds are lacking for peakVO 2 , hand-grip strength, and general fatigue, strong conclusions about HI's cost-effectiveness as compared to LMI exercise for these outcomes cannot be made.
Strengths of this study include the direct comparison between HI and LMI exercise interventions, longer-term effectiveness and cost-effectiveness analyses, multicentre RCT design, large sample size, the use of valid and reliable outcome measures, and the use of state-of-the-art statistical methods. However, some limitations are noteworthy. First, to limit nonparticipation and minimize contamination, a WLC was included instead of a non-exercising control group. Therefore, at longer term, we were only able to evaluate the effectiveness and cost-effectiveness of HI compared to LMI exercise, because all participants had received an exercise intervention at 64 weeks. Second, cost data were collected using self-report, which may have caused Bsocial desirability^and/or Brecall bias.^Third, a relatively large number of participants had missing cost data. To deal with this limitation, missing data were multiple imputed [42]. Finally, it should be acknowledged the CEA results might not be generalized to other countries with different healthcare systems and/or payment structures [43].

Conclusions
In conclusion, at longer-term follow-up, we found a larger intervention effect on role and social functioning for HI than for LMI exercise. Exercise-induced benefits in peakVO 2 and HRQoL were successfully maintained between short and longer terms, but not for fatigue. Furthermore, HI exercise was cost-effective for QALYs compared to LMI exercise, mostly due to significant lower healthcare costs in HI exercise. Hence, the current findings advocate the implementation of supervised exercise as part of standard cancer care, and if possible HI exercise.
Funding This work was supported by the Alpe d'HuZes/KWF Fund. The research grant is provided by the Dutch Cancer Society (Grant Number ALPE2009-4619).

Compliance with ethical standards
Ethical approval This study is approved by the Medical Ethics Committee of the VU University Medical Centre (Amsterdam, The Netherlands).