Cost-effectiveness of prehabilitation prior to elective surgery: a systematic review of economic evaluations

Prehabilitation is still a relatively new care concept. It aims at enhancing patients’ functional capacity and overall health status through behaviour change [1] to enable them to withstand a forthcoming stressor [2]. In the surgical context, prehabilitation complements the concept of ‘enhanced recovery after surgery’ (ERAS) and aims to improve surgical outcomes and lower post-operative complication rates [3]. Prehabilitation programmes are delivered preoperatively by a multidisciplinary team and in various settings (e.g. inpatient, outpatient, or at home). Typical modalities include exercise training, promotion of physical activity, nutritional optimisation and psychological support [4], which are provided in addition to elements of ERAS, such as medical optimisation and alcohol or smoking cessation [5].

The potential of prehabilitation is widely recognised. Nevertheless, prehabilitation has not yet been widely adopted by health care systems. Current evidence is still somewhat limited, though much research is still underway to determine the optimal programme types and delivery modalities for different patient populations. Most research activity seems to be in the field of cancer surgery, for example, in an overview of 55 systematic reviews on preoperative prehabilitation, 23 reviews specifically focused on cancer [6]. A likely explanation for this phenomenon is that there is already a large body of evidence demonstrating the positive effects of physical activity on the physical and psychological outcomes of cancer patients [7]. In addition, little is known about the cost-effectiveness of prehabilitation, which is critical for policy-makers considering the implementation of such programmes. By definition, prehabilitation is an approach to reduce healthcare costs [4] and a comprehensive analysis of the value of prehabilitation should incorporate cost outcomes [8].

The aforementioned overview identified only one systematic review on costs [6], but this review focused on nutritional support rather than full prehabilitation programmes [9]. Other reviews that addressed health economic outcomes focused on specific populations [10] or were not systematic reviews [11]. One large systematic review including 178 randomised controlled trials (RCTs) showed that prehabilitation may reduce postoperative length of stay and complications [12], both of which would translate into a cost reduction. However, to our best knowledge, there is currently no comprehensive systematic review on the cost-effectiveness of prehabilitation prior to elective surgery.

The aim of this systematic review was to synthesise the evidence on the cost-effectiveness of prehabilitation programmes for patients awaiting elective surgery compared with usual preoperative care to inform decisions about the implementation of prehabilitation programmes and to guide the design of future rigorous economic evaluations of prehabilitation programmes. More specifically, our objectives were to (1) identify all eligible economic evaluations (EEs), (2) assess their validity and (3) systematically present their characteristics, methods and findings.

The systematic review was prospectively registered in PROSPERO (CRD42020182813) and we published a protocol [21]. Important protocol changes are reported in Additional file 1: Appendix 1.

The study in- and exclusion criteria are displayed in Table 1 (from the protocol with additional specifications) [21].

We searched PubMed, Embase and the Centre for Reviews and Dissemination (CRD) Database on 31/08/2021, which are the most efficient combination of bibliographic databases for systematic reviews of EEs [22]. Furthermore, we searched OADT.org and the DART-Europe E-theses Portal for grey literature and ClinicalTrials.gov and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) for unpublished and ongoing studies on October 30, 2021. A weekly email alert was created for the search in PubMed (monitored until August 23, 2022). Additionally, we screened the reference lists of included EEs and relevant systematic reviews as well as articles citing the included EEs obtained through Web of Science and Google Scholar. We also contacted the corresponding authors of all included EEs about further relevant EEs.

The database search strategies consisted of search terms, relating to the population (e.g. ‘preoperative’), the intervention (e.g. ‘exercise’) and study type, i.e. terms to search for economic evaluations (e.g. ‘cost’). Full search strategies for all sources can be found in Additional file 1: Appendix 2.

Records retrieved from databases were deduplicated, screened and managed using EndNote 20 (Clarivate Analytics, Philadelphia (PA), USA). After deduplication, a randomly selected 10% sample of all unique records was screened against the eligibility criteria by two reviewers (TR, HE) independently based on their titles and abstracts. Disagreement was resolved by consensus. As agreement was above 80%, the remaining 90% were screened by one reviewer (TR). We retrieved the full-text articles for all potentially eligible studies as well as for relevant systematic reviews, so that their references could be screened. Each full-text article was screened for eligibility by two reviewers independently who noted reasons for exclusion. Disagreements were resolved by consensus and by consulting a third reviewer (WQ). Last, all study reports were mapped to unique studies as the unit of interest. No automation tools were used in the process.

Data were extracted into a standardised excel sheet that was piloted by one reviewer (TR). Two reviewers (TR, HE) independently extracted the data of a randomly selected 20% sample of the included completed EEs for calibration. Disagreement was resolved by consensus. As there were no systematic discrepancies, the remaining records were extracted by one reviewer (TR). All outcome data was verified by a second person (JS). We used all documents relevant to the included EEs for data extraction and contacted the study authors via email in case of missing or unclear data. Uncertainties about the methods were only inquired for completed EEs. A reminder email was sent after 2 weeks.

A list of all data items and detailed descriptions can be found in Additional file 1: Appendix 3. For ongoing studies, we only extracted the study characteristics and, if published as a protocol, the EE methods. For completed EEs, we also extracted post-operative results data (per group and as the difference between groups) on clinical effectiveness and costs. Costs were reported with their original year and currency as well as converted to 2020 EUR. For conversion, we used the ‘Cochrane Campbell Economic Methods Group and the Evidence for Policy and Practice Information and Coordinating Centre Cost Converter’ (version 1.6) [23]. We only extracted unadjusted data for the last available follow-up point based on intention-to-treat analyses.

The risk of bias of trial-based EEs was assessed on outcome-level using the Cochrane risk of bias tool 2 (RoB 2) [24] for EEs based on RCTs and the ROBINS-I tool [25] for EEs based on non-randomised studies of interventions (NRSI). Among other domains, both tools address the risk of reporting bias. Risk of bias figures were created for each outcome domain separately using the robvis application [26]. Methodological quality of trial-based full EEs was assessed using the Consensus on Health Economic Criteria (CHEC) checklist [27]. Model-based EEs were assessed for credibility using the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) checklist [28]. Assessments were performed by two reviewers (TR, HE) independently in a random 20% sample of the included EEs and continued by one reviewer (TR) as agreement was above 80%.

The review’s primary outcome was the cost-effectiveness from cost–utility analyses (CUAs) based on direction of effect (i.e. reduced costs and/or additional quality-adjusted life year gained). Secondary outcomes were the cost-effectiveness from cost-effectiveness analyses (CEAs), cost–benefit analyses (CBAs), cost-minimisation analyses (CMAs) and cost–consequence analyses (CCAs) based on direction of effect. We calculated effect measures when not reported using risk differences for dichotomous outcomes and mean differences or differences in medians for continuous outcomes. Confidence intervals were extracted when reported. All calculated values are marked as such. All outcomes were reported in disaggregated form in natural units and combined outcome measures, e.g. incremental cost-effectiveness ratios (ICERs), where possible.

We were unable to perform a meta-analysis because the only EEs that were sufficiently homogenous had an unquantifiable overlap in patient populations [29‐34] or missed crucial information for data transformation [32, 33]. Therefore, structured narrative synthesis in the form of vote counting based on direction of effects was performed [35]. EEs were grouped by design (model-based vs. trials-based) [16] and analysis type (CUA vs. CEA, CBA, CCA, CMA) to reflect the prioritisation of outcomes.

Results were presented graphically in form of a hierarchical permutation matrix [36]. There were ten possible outcomes for incremental costs (which could be higher, lower or same) and effectiveness (which could be better, poorer, same or inconsistent) corresponding to five result categories: cost-effective, neutral, not cost-effective, unclear; incremental analysis required, and unclear; individual decision required). No formal sensitivity analysis was performed but we discussed the influence of excluding EEs that were of critical risk of bias or insufficient credibility. Descriptive post-hoc subgroup analyses were performed to explore heterogeneity in the EEs’ results arising from differences in populations, interventions, methods, funding source and conflict of interest.

To address publication bias, we searched comprehensively for ongoing studies and grey literature and followed up on their status by searching for related publications and contacting the named investigators. In addition, we discussed how the effectiveness results from the included EEs compare to those of clinical effectiveness studies on prehabilitation using an overview of 55 systematic reviews and meta-analyses of RCTs by McIsaac et al. 2022 [6]. Our hypothesis was that the EEs would appear more beneficial if there truly was a publication bias.

The study selection process is presented in Fig. 1. In total, 45 unique studies were included: 25 completed EEs [29‐34, 37‐55] and 20 ongoing studies, of which 11 were published as protocol articles [56‐66] and nine as registration records [67‐75]. Two completed EEs were only published as conference abstracts [53, 54] and two as dissertations [37, 49]. A total of 54 email enquiries were sent to the study authors, of which 23 were answered (response rate 42.6%). A list of all articles excluded after full-text screening can be found in Additional file 1: Appendix 4, with an additional explanation for close misses and articles excluded post hoc [76‐85].

The characteristics of the 25 completed EEs are displayed in Table 2. In summary, there were 22 trial-based EEs (13 RCTs and 9 NRSI), and three model-based EEs (2 decision trees and 1 financial projection) corresponding to four CUAs, three CEAs, two CBAs, 12 CCAs and four CMAs. Nine EEs were performed from a mix of a payer and provider perspective, three EEs each from a payer or provider perspective, and one EE from a patient perspective. The perspective remained unclear in the remaining nine EEs.

Most EEs were published in the last 10 years and came from Europe (10 EEs), Asia (8 EEs) or North America (7 EEs). The EEs covered a wide range of diseases and surgery types that can be broadly categorised as orthopaedic surgery (9 EEs), cancer surgery (8 EEs), mixed major surgery (6 EEs) and other (2 EEs). In 11 EEs, patients had an increased perioperative risk (e.g. old age or frailty). Sample size ranged from 20 to 8830 patients (median 137). The median proportion of women across the EEs was 53%, and the mean or median age ranged from 59 to 78 years, with one outlier (median age 27 years).

Characteristics and methods of the 20 ongoing EEs are reported in Additional file 1: Appendix 5. All are trial-based EEs, with the majority based on RCTs (18 EEs). There were five CUAs, six CEAs, three EEs using both CUA and CEA, and six EEs with unclear analysis type. In addition to the above continents, two ongoing EEs were from Australia and one from South America. The disease and surgery types were similar to the completed EEs (9 EEs on cancer, 8 EEs on major mixed or major other surgeries and 3 EEs on orthopaedics), though there were slightly more EEs from the field of cardiology and focusing on patients with an increased perioperative risk, and less EEs from the field of orthopaedics.

Information on the completed and ongoing EEs’ funding and conflict of interest can be found in Additional file 1: Appendix 6. Nine EEs did not report any information, one received parts of its funding from a commercial funder [57], one from a private donor [46] and in one, it was unclear [68]. Two EEs declared a relevant conflict of interest [29, 42], as authors were related to companies contracted to organise the prehabilitation.

Detailed information on the methods can be found study-by-study in Additional file 1: Appendix 7 (completed EEs) and Additional file 1: Appendix 8 (ongoing EEs published as protocols). Most completed EEs used a time horizon for effects and costs of 1 month or less (range: 2 weeks to 24 months), with various EEs following patients until discharge and using the costs of hospital stay. No EE discounted effects or costs. Using bootstrapped precision measures (e.g. 95% confidence intervals) was the most common method for calculating uncertainty around the point estimates. Three EEs applied willingness-to-pay thresholds. In summary, with two exceptions [40, 42], few EEs applied comprehensive economic evaluation methods.

Characteristics of the completed EEs’ prehabilitation programmes can be found in Table 3. Briefly, in most EEs, the prehabilitation programme was multimodal. All 25 programmes included an exercise element, though the type of training and use of unsupervised sessions varied. Additionally, many included an element of counselling or education (13 EEs) or an element addressing the patients’ nutritional status (11 EEs). The programmes involved various groups of health care professionals, the most common group being physiotherapists (14 EEs). Most programmes were performed in an outpatient (11 EEs) or home setting (8 EEs). The programmes’ overall duration ranged from 3 days to 3 months, with most programmes lasting between 2 and 4 weeks. The frequency of supervised sessions ranged from daily to once per week, with session durations being individual or ranging from 30 to 70 min. Where intensity was reported, we mostly classified it as high, e.g. an 80% of peak work rate for endurance training. Many programmes were not evidence-based. They costed between 100 and 1000 EUR (2020) per patient. Characteristics of the programmes evaluated in the ongoing studies can be found in Additional file 1: Appendix 9.

The results of the assessment of risk of bias and methodological quality of the included studies can be found in Additional file 1: Appendix 10. The majority of RCT-based EEs were judged to be of high risk of bias with the RoB 2 tool. Only one RCT had a moderate risk of bias in all domains [42], and none had a low risk of bias. The main reason for high risk of bias was the absence of a prospective study protocol/registration record. All NRSI-based EE had at least a high, one even a critical risk of bias [34], the main reason being that most EEs did not adequately control for confounding when selecting or analysing patients.

The methodological quality of full trial-based EEs as judged with the CHEC-checklist ranged from 8 to 15 fulfilled items (of 18 to 19 applicable items) and thus can be considered moderate to low. The credibility of model-based EEs as judged with the ISPOR checklist was acceptable in one EE [40], insufficient in another EE [41] and could not be determined due to lack of information in one EE published as a conference abstract [54].

Table 4 provides an overview of the results of the completed EEs. Smaller values represent a higher benefit unless indicated otherwise. Morbidity refers to the rate of postoperative complications unless indicated otherwise. Detailed cost results can be found in Additional file 1: Appendix 11 including quantities of resource use, unit costs, total costs, incremental cost-effectiveness ratio (ICER) in the original currency and year, and the study authors’ conclusion. Furthermore, adherence and safety/feasibility outcomes can be found in Additional file 1: Appendix 12. Two EEs had adherence rates of less than 35% [48, 51] and in three EEs, drop-out and/or adverse event rates were notably higher in the prehabilitation groups [47, 48, 53].

Four trial-based EEs [37, 42, 48, 53] reported data on the primary outcome, i.e. cost-effectiveness based on CUA (Fig. 2, thick bordered column). Based on direction of effects, three CUAs (75%) fell into the cost-effective category, and one fell into the category ‘unclear; incremental analysis required’. The ICER of the latter study was 7906 EUR (2020) per quality-adjusted life year (QALY) gained, which is likely acceptable under common willingness-to-pay (WTP) thresholds [37].

Three model-based and 18 trial-based EEs reported on the secondary outcomes (Fig. 2), i.e. cost-effectiveness based on other types of EEs, respectively. Based on direction of effects, two model-based EEs fell into the cost-effective category [40, 41], but one was judged insufficiently credible [41]. The remaining model-based EE fell into the category ‘unclear; individual decision required’ [54]. Of the trial-based EEs, 11 fell into the cost-effective category [29, 31‐34, 38, 46, 47, 49, 50, 55], one of which was judged to be of critical risk of bias [34], three into the category ‘unclear; incremental analysis required’ [43, 51, 52], three into the category ‘unclear; individual decision required’ [30, 44, 45], and one, a CMA with a difference in total costs of + 2 EUR (2020), into the not cost-effective category [39].

Overall, 16/25 (64.0%) EEs found prehabilitation cost-effective based on direction of effects, (14/23; 60.9% when excluding the EEs of insufficient credibility/critical risk of bias [34, 41]), in 8/25 (32.0%) it was unclear, and one EE (4.0%) found prehabilitation not cost-effective [39]. Descriptive post hoc subgroup analyses revealed heterogeneity in the cost-effectiveness results depending on the population, intervention and methods, but not on conflict of interest and funding source (Additional file 1: Appendix 13). Briefly, cost-effectiveness was more frequently observed in EEs of cancer patients, patients with a high perioperative risk, multimodal programmes, home-based or inpatient prehabilitation, shorter programmes, low-cost programmes and EEs taking a mix of payer/provider perspective.

There was a relevant risk of publication bias regarding the included completed EEs. Firstly, the review had initially included 74 study reports belonging to 54 unique studies. However, ten reports referring to nine unique studies were excluded post hoc [76‐85] (see Additional file 1: Appendix 4), of which four protocols [77‐80], one registration record [85] and a conference abstract [81] referred to studies that no longer reported on costs in the study publication [87‐95]. The authors of two studies confirmed that no economic evaluation was performed [78, 80]. The remaining authors did not respond.

In comparison to the results of the overview by McIsaac et al. 2022 [6], the included EEs on cancer surgery showed more beneficial results regarding morbidity [32‐34, 43, 46, 55] and mortality [41, 50, 55], and the included EEs on orthopaedic surgery showed more beneficial results on health-related quality of life (HrQoL) [37, 42, 47, 48, 53]. Apart from that, results were comparable but, overall, the included EEs’ results appear more beneficial suggesting a risk of publication bias.

Some limitations on review and study level apply. First, we could not perform a meta-analysis but had to resort to narrative synthesis in the form of vote counting based on direction of effects. This synthesis method does not provide any information on the magnitude of effects, nor does it account for the EEs’ sample sizes [106]. Second, the review’s broad inclusion criteria led to a large number of included articles that we coined ‘EEs’ for the purpose of the review. However, most of them were trial reports including cost outcomes which understood themselves as pilot and/or feasibility trials and thus did not apply comprehensive EE methods. Third, as there currently is no universally recognised definition of prehabilitation [6] nor common concepts, procedures or measurements [4], the definition of the prehabilitation elements varied between the EEs. The definition of usual care also varied across EEs. For instance, advice on physical activity and smoking cessation were included as standard care in some EEs [38, 50, 51], while those aspects were part of prehabilitation in other EEs [29‐31, 40, 41, 54]. Lastly, characteristics of health systems, such as the type of financing (public vs. private) and organisation of care (centralised vs. decentralised), play a crucial role in programme delivery and cost justification. As we did not formally assess the generalisability and transferability of our results to different health systems, we recommend policy-makers interested in implementing prehabilitation to conduct a health technology assessment (HTA) for their government.

Limitations on the study level included the high risk of bias and low methodological quality of the included EEs. However, the exclusion of the two EEs judged to be insufficiently credible/of critical risk of bias only had a small effect on the results. Furthermore, there was a high risk of publication bias associated with trial-based EEs. Trial-based EEs are by nature prone to a specific form of publication bias, namely conduct bias [107], meaning they are not published because they were never performed in the first place, e.g. when the underlying trial was ‘inconclusive’ or had negative results. Although an intervention that is less effective but cheaper than the control may still be cost-effective, it is generally not acceptable from an ethical and quality of care perspective to replace usual care with a less-effective intervention.

Owing to the limitations described above, our results should be interpreted with caution. As many EEs were based on prospective trials, decision-makers must also consider the possibility that there was a motivational bias among the participants and that the cost-effectiveness of prehabilitation may be lower under ‘real world’ circumstances. Before implementing prehabilitation into routine care, decision-makers should assess potential barriers and facilitators [108, 109], which may differ between health systems and stakeholders, or even individuals. For example, qualitative studies found that group prehabilitation was perceived both as a barrier and facilitator [110, 111]. In their framework for prehabilitation services, Bates et al. 2020 list several considerations for the implementation of prehabilitation, including to involve patients when designing the prehabilitation programme [112].

Finally, decision-makers must determine which patient population(s) should receive prehabilitation and establish screening pathways, accessibility to the programme and strategies to ensure sustainability [113]. This involves performing a budget impact analysis, including the one-time investments into infrastructure (e.g. prehabilitation centres) as well as the running costs for the provision of prehabilitation and maintenance of the infrastructure. Although many EEs found that prehabilitation paid off during the index hospitalisation, the pervasive shortage of health care professionals [114] may hinder implementation of prehabilitation.

First, future research should address the knowledge gaps discussed above, i.e. which populations benefit most and what the optimal prehabilitation programme for those populations is. If a broadly defined population is included in a clinical trial, it is recommended to consider pre-specified subgroup analyses for economic evaluation [115]. To ensure added value, new clinical research should consider the existing evidence [116] as well as involve patients and stakeholders in all phases of research [117], e.g. when designing the prehabilitation programme [118]. Ideally, these efforts would result in a clinical practice guideline for prehabilitation, the first step of which was taken by Tew et al. 2018 with a guideline on preoperative exercise training in patients awaiting major noncardiac surgery [119].

Second, future research should address the shortcomings of existing EEs. Common issues included inadequate reporting, short time horizons, and the use of limited perspectives. Reporting guidelines are intended to support authors and increase the accuracy and transparency of reporting, but they are frequently used inappropriately, including those for EEs [120]. In our review, reporting guidelines for EEs seemed to have been under-used, as none of the full EEs published as full-text articles after 2013 [37, 40‐42, 48, 52] reported following the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) [121], which is applicable to both trial- and model-based EEs. A possible reason is that only two trial-based EEs were published as separate full-text articles [38, 42]. Hence, we recommend that authors publish full EEs as separate articles and follow the latest version of the CHEERS checklist [122].

Many EEs had a short time horizon of 1 month or less. However, as argued by Grocott and Ludbrook 2019, ‘it is plausible that improved fitness arising from prehabilitation might have a further lingering positive impact on the need for later care’ [123]. Such an impact can only be detected using a longer time horizon but, in our review, only five EEs [39, 41, 42, 48, 53] had a time horizon of 12 months or more. To determine an adequately long-time horizon, we recommend authors to consult guidelines from their national HTA institutes and by the ISPOR [115]. On a closely-related matter, many EEs applied limited perspectives, such as the provider perspective, with the hospital being the provider, and therefore did not consider post-discharge or out-of-hospital resource use. None of the included EEs applied a full societal perspective including costs from other sectors, e.g. productivity loss. When this is not feasible, we suggest that authors adopt a comprehensive health sector perspective including all relevant payers and providers. For example, EEs may consider improved access for other patients through freed-up capacity, e.g. due to earlier discharge of prehabilitated patients [124]. In summary, future EEs should be performed over a longer time horizon and apply a more comprehensive perspective.

We plan to update the review upon publication of our own economic evaluation [59] in 2025/26 by re-running the search strategies modified only by adding the MeSH/Emtree term ‘Preoperative Exercise’.