Patient decision support interventions for candidates considering elective surgeries: a systematic review and meta-analysis

Background: The increase in elective surgeries and varied postoperative patient outcomes has boosted the use of patient decision support interventions (PDSIs). However, evidence on the effectiveness of PDSIs are not updated. This systematic review aims to summarize the effects of PDSIs for surgical candidates considering elective surgeries and to identify their moderators with an emphasis on the type of targeted surgery. Design: Systematic review and meta-analysis. Methods: We searched eight electronic databases for randomized controlled trials evaluating PDSIs among elective surgical candidates. We documented the effects on invasive treatment choice, decision-making–related outcomes, patient-reported outcomes, and healthcare resource use. The Cochrane Risk of Bias Tool version 2 and Grading of Recommendations, Assessment, Development, and Evaluations were adopted to rate the risk of bias of individual trials and certainty of evidence, respectively. STATA 16 software was used to conduct the meta-analysis. Results: Fifty-eight trials comprising 14 981 adults from 11 countries were included. Overall, PDSIs had no effect on invasive treatment choice (risk ratio=0.97; 95% CI: 0.90, 1.04), consultation time (mean difference=0.04 min; 95% CI: −0.17, 0.24), or patient-reported outcomes, but had a beneficial effect on decisional conflict (Hedges’ g=−0.29; 95% CI: −0.41, −0.16), disease and treatment knowledge (Hedges’ g=0.32; 95% CI: 0.15, 0.49), decision-making preparedness (Hedges’ g=0.22; 95% CI: 0.09, 0.34), and decision quality (risk ratio=1.98; 95% CI: 1.15, 3.39). Treatment choice varied with surgery type and self-guided PDSIs had a greater effect on disease and treatment knowledge enhancement than clinician-delivered PDSIs. Conclusions: This review has demonstrated that PDSIs targeting individuals considering elective surgeries had benefited their decision-making by reducing decisional conflict and increasing disease and treatment knowledge, decision-making preparedness, and decision quality. These findings may be used to guide the development and evaluation of new PDSIs for elective surgical care.


Introduction
The global increase in elective surgeries [1,2] and varied postoperative patient outcomes are growing public health concerns [3] . Undesirable patient outcomes such as postoperative complications [3] and decreased health-related quality of life (HRQoL) [4] are common. Therefore, surgical decisions should be 'preference-sensitive' [5] ,

HIGHLIGHTS
• Patient decision support interventions (PDSIs) had no effect on invasive treatment choice, consultation time, or patient-reported outcomes (PROs). • PDSIs had a beneficial effect on decisional conflict, disease and treatment knowledge, decision-making preparedness, and decision quality. • Treatment choice varied with surgery type.
• Self-guided PDSIs had a greater effect on disease and treatment knowledge enhancement than cliniciandelivered PDSIs.
namely, guided by patient preferences when several options are available or patient outcomes are uncertain. PDSIs have been used to enhance the preference-sensitive nature of clinical decision-making. PDSIs present evidence-based information to patients about a health condition, treatment options, and the associated benefits and risks, and implicitly or explicitly clarify the value patients place on the treatment benefits and risks [5] . The primary goal of PDSIs is to enhance decisionmaking quality and facilitate patient engagement during consultations [6] . PDSIs may also assist patients by increasing their knowledge of the available options and outcomes, thereby equipping them with more realistic expectations [6] .
With the increasing availability of validated PDSIs for elective surgical candidates faced with a treatment decision [7][8][9] , it is essential to investigate the effects for various types of elective surgery. In general, PDSIs have been found to improve disease and treatment knowledge [5,8,9] , satisfaction with decision-making [9] , decision quality [8] , and reduce decision conflict [5,8,9] . However, these reviews included different types of PDSIs [8,9] , study designs [8,9] , and elective surgery is quite unique.
Knops et al. [7] concluded that PDSIs increased knowledge and decreased decision conflict, but had no effect on anxiety and postoperative HRQoL. However, the review is outdated [7] . Therefore, we updated this review with the aims of (1) summarizing the effects of PDSIs on invasive treatment choice, decisionmaking-related outcomes, PROs, and healthcare resource utilization outcomes for patients considering elective surgery, and (2) identifying the moderators of PDSIs effects, with an emphasis on the type of targeted surgery.

Protocol and registration
This systematic review and meta-analysis were conducted according to the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines (Supplementary Table 1, Supplemental Digital Content 1, http://links.lww.com/JS9/ A238) [10] . The protocol was registered in the International Prospective Register of Systematic Reviews database (PROSPERO Number: CRD42021273767).

Eligibility criteria
The eligibility criteria is illustrated in Supplementary Table 2 (Supplemental Digital Content 1, http://links.lww.com/JS9/ A238). We considered all types of randomized controlled trials (RCTs), including both published and unpublished trials. The trials had to enroll surgical candidates who were contemplating elective surgeries, defined as a surgical procedure that is scheduled in advance because it does not involve a medical emergency. The included trials must have assessed one or more PDSIs, defined as tools designed to inform patients and clinicians of their elective treatment options, surgical or both surgical and nonsurgical, among which none is the undisputable choice to all patients. They could take the form of computer software or physical tools. Comparators had to include an active control group, a standard care group, or a waitlist group. Outcomes had to include invasive treatment choices, decision-making-related outcomes, PROs, and outcomes related to healthcare resource use. The trials were limited to studies published in the English language, but not publication period. We excluded trials in which individuals had cognitive impairment or psychiatric disease.

Information sources and search strategy
A three-step comprehensive search strategy was employed from inception to 30 August 2021, under the guidance of an experienced librarian team. First, we searched eight databases (search engines) (index and key terms provided in Supplementary Table 3 Table 4, Supplemental Digital Content 1, http://links.lww.com/JS9/A238) for relevant ongoing and unpublished trials. Third, we conducted a manual search on the reference lists of the related primary studies or systematic reviews. In addition, we searched specialized journals and grey literature databases for potential trials. The EndNote X20 software was used to manage the references and excluded duplicates [11] .

Study selection and data extraction
Study selection was graphically illustrated using the PRISMA 2020 flow diagram (Fig. 1). Two reviewers (L.J.C. and M.X.L.) independently screened the titles and abstracts to identify their relevance. When multiple reports of the same study were identified, studies were collated. The potential full texts were selected based on the eligibility criteria, and the reasons for inclusion/exclusion were recorded (Supplementary Table 5, Supplemental Digital Content 1, http://links.lww.com/JS9/A238). Any disagreements were resolved by a third reviewer (N.L.).
A standardized data extraction form was developed using the Cochrane Handbook for Systematic Reviews of Interventions [12] . Three essential components were included: trial characteristics, PDSIs characteristics, and key outcomes related to decisionmaking, PROs, and healthcare resource use. Based on a reporting guide [13,14] , the following characteristics of PDSIs were extracted: aim, element, platform, co-intervention, duration, media format, use of PROs data, artificial intelligence embodiment, value consideration, theoretical framework, communication, facilitator, survey administration methods at various time points, and assessment intervals.
Invasive treatment choice included the actual choice of invasive treatment implemented; if not specified, the participant's preferred option was used as a surrogate measure [7] . Decisionmaking-related outcomes (measures) included decisional conflict (different versions of Decisional Conflict Scale [15] ), satisfaction with decision-making (effective decision subscale of Decisional Conflict Scale and self-developed patient satisfaction surveys), disease and treatment knowledge (self-developed questionnaires), decisional regret (Decision Regret Scale [16] ), Preparedness For Decision-Making (Rochester Participatory Decision-Making Scale [17] and Preparedness For Decision-Making Scale [18] ), decision quality (Decision Quality Instrument [19] ), Shared Decision-Making (CollaboRATE [20] and Nine-Item Shared Decision-Making Questionnaire [21] ), Decision Self-Efficacy (Self-Efficacy For Managing Chronic Diseases Six-Item [22] and Decision Self-Efficacy Scale [23] ), and outcome expectations.
PROs included HRQoL (general or condition-specific), physical health, mental health, depression, anxiety, and perceived stress. Outcomes related to healthcare resource use included consultation time. Detailed definitions for all extracted outcomes and the measures used are listed in Supplementary Table 6 (Supplemental Digital Content 1, http://links.lww.com/JS9/A238).
The data extraction form was piloted on ten trials to ensure that the items were accurate and appropriate. Two independent reviewers (L.J.C. and M.X.L.) retrieved data from 58 trials following item confirmation. If trials reported data on the median, range, interquartile range, and SE, data conversion was utilized to compute the mean and SD using recommended formulae [12,24,25] . When data was questionable, insufficient, or missing, trial authors were contacted via email and asked to provide additional unpublished details or results.

Quality of patient decision support interventions design
International Patient Decision Aid Standards instrument shortform (IPDASi-SF) was used to assess the quality of PDSIs [26] . The IPDASi-SF contains 16 items addressing seven dimensions related to the information about the options, probabilities, values, development, disclosure of funding, decision aid evaluation, and evidence used [26] . The IPDASi-SF score ranged from 0 to 16, with a higher score indicating better PDSIs quality.

Risk of bias in individual studies
The Cochrane risk of bias (RoB 2.0) [27] tool was used to assess the risk of bias in the selected randomised trials on five domains of bias: bias arising from the randomization process, bias due to deviations from the intended intervention, bias due to missing outcome data, bias in the measurement of the outcome, and bias in the selection of the reported result. Two reviewers (L.J.C. and M.X.L.) independently responded to each of the signaling questions with (1) yes, (2) probably yes, (3) probably no, (4) no, or (5) no information. The RoB 2.0 algorithmic tool rates the risk of bias as (1) low risk of bias, (2) some concerns, or (3) high risk of bias for each domain.

Statistical analysis
The Meta command procedures in Stata 16 software were used to perform a meta-analysis and subgroup analysis [28,29] . Z-statistics with a significance level of P-value less than 0.05 was used to assess the overall effect [30] . A minimum of two studies was needed to perform a meta-analysis [31] . A weighted risk ratio (RR) (dichotomous data), standardized effect size, or mean difference (continuous data) was calculated for each outcome measure [32] . As most of the trials used a small sample size, the pooled effect sizes of outcomes (continuous data) were assessed using Hedges' g [33] . The effect magnitude was classified as small ( ≥ 0.2), medium ( ≥ 0.5), large ( ≥ 0.8), or very large ( ≥ 1.2) [34] . We used the DerSimonian and Laird procedure to estimate the variability between studies for random-effects meta-analysis [35] . Statistical heterogeneity was assessed using I 2 statistics and Cochran's Q test, with a P-value less than 0.10 indicating evidence of heterogeneity [36] . The degree of heterogeneity using overlapping intervals for I 2 was set as 0-40% (might not be important), 30-60% (moderate), 50-90% (substantial), and 75-100% (considerable) [37] . The source of heterogeneity was investigated using subgroup analysis [37] . For each outcome measure, whenever 10 or more trials were available, separate subgroup analysis was performed to evaluate five moderators coded into categorical variables including: types of elective surgery, patient-reported outcome measures-based PDSIs (yes/no), mode of delivery (self-guided/clinician-administered), value consideration (yes/no), and use of theoretical framework (yes/no).

Reporting bias assessment
We used Begg's test [38] , Egger's regression test [39] , the asymmetry of the funnel plot [40] , and the trim-and-fill approach [40] to examine publication bias in meta-analyses with 10 or more trials [41] . Egger regression and Begg's test were performed using Stata 16 software [28] , with a P-value more than 0.05 indicating no small-study effects existed. We identified the possibility of heterogeneity in effect sizes across studies, limiting the conclusions drawn from Egger's tests and the funnel plot [42,43] . We also applied the Copas selection model, using the metafor package in R software [44] , to account for selection bias according to funnel plot asymmetry [45,46] .

Certainty assessment
GRADEpro 3.6 software [47] was used to assess the certainty of evidence using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach [48] . The GRADE assessment focuses on five factors: methodological limitations, imprecision, inconsistency, indirectness, and publication bias [49] . When issues were detected in the five factors, the evidence was downgraded. The certainty of the evidence was graded as high, moderate, low, and very low. Figure 1 illustrates the study selection results. We identified 6142 records and eliminated 1070 duplicates. In addition, citation searching yielded 14 records and 15 studies, as well as two reports, from a previous review [7] . After two reviewers screened independently, 4990 records were excluded based on their titles and abstracts. Eighty-two full-text articles were selected, and 50 were excluded for several reasons (Supplementary Table 5, Supplemental Digital Content 1, http://links.lww.com/JS9/ A238). Finally, 58 trials  were included in the review, with 52 unique trials and six reports linked to the unique trials (Supplementary Table 7, Supplemental Digital Content 1, http:// links.lww.com/JS9/A238). The inter-rater agreement of the two reviewers was consistent (κ = 0.87, P < 0.001).

Trial characteristics
The characteristics of trials, published between 1995 and 2021, are summarized in Table 1. The sample size ranged from 16 to 1485 participants, with a mean age between 23.6 and 72.0 years. The attrition rate varied between 0% and 44.8%. Of these, 52 trials (89.7%) compared PDSIs with usual care, with the remaining six trials comparing different PDSIs. Thirty-two trials (55.2%) were conducted in the USA, with seven in Canada (12.1%), six in The Netherlands (10.3%), three in Australia (5.2%), three in the UK (5.2%), two in Finland (3.4%), and only one in China, Germany, Hong Kong, Spain, and Turkey. Fiftytwo trials (89.7%) utilized a two-arm RCT design, two trials used a three-arm RCT and a cluster RCT design, and one trial used a stepped wedge trial design. Most of the trials studied PDSIs developed for patients with neoplasms (number of trials, k = 25, 43.1%), diseases of the musculoskeletal system (k = 18, 31.0%), or diseases of the genitourinary system (k = 8, 13.8%). Most of the PDSIs evaluated involved decisions about elective gynecology and obstetrics surgery (k = 21, 36.2%) or orthopedic surgery (k = 14, 24.1%). The evidence on the quality of PDSIs was available for all 58 trials (Supplementary Table 9, Supplemental Digital Content 1, http://links.lww.com/JS9/A238). Eleven PDSIs met all IPDASi-SF criteria, and their total score ranged from 5 to 16 (median = 12). Full information on the available options, their positive and negative features, and fair comparisons, was provided for all PDSIs. However, incomplete information was provided regarding the impartial reviews (33 of 58 PDSIs; 56.9%), citations to referenced studies (32 of 58 PDSIs; 55.2%), testing details with patients (29 of 58 PDSIs; 50%), and production date (24 of 58 PDSIs; 41.4%). Ten trials incorporated the Ottawa Decision Support Framework (ODSF) to inform the design of the decision assistance, out of 18 trials that incorporated the theoretical framework.
There were no significant changes in the following decisionmaking-related outcomes: decisional regret, shared decisionmaking, decision self-efficacy, and outcome expectations ( Table 2).

Publication bias
Publication bias was not detected for treatment choice, patient satisfaction, and disease and treatment knowledge; however, there was an asymmetrical distribution on the funnel plots for decisional conflict (Supplementary Figs. 11a, c, Fig. 12, Supplemental Digital Content 1, http://links.lww.com/JS9/ A238). A sensitivity analysis using the Copas selection model suggested that publication bias was unlikely to be an issue ( Supplementary Fig. 13, Supplemental Digital Content 1, http:// links.lww.com/JS9/A238).

Certainty of evidence
Supplementary Table 10 (Supplemental Digital Content 1, http:// links.lww.com/JS9/A238) shows the GRADE summary of evidence. For all primary and secondary outcomes, the certainty of evidence was rated as very low or low, except for disease and treatment knowledge (follow-up), decision self-efficacy, physical health, anxiety (follow-up), depression (postintervention), and consultation time, which were rated as moderate, while mental health and depression (follow-up), as well as perceived stress, which were rated as high.

Summary of evidence
Our review demonstrated PDSIs that were intended to guide decision-making for elective surgeries had a beneficial impact on many decision-related outcomes. These effects were small and  Moderate g, Hedges's g; HRQoL, health-related quality of life; I 2 , percentage of variation across studies that is due to heterogeneity rather than chance; MD, mean difference; PDSI, patient decision support intervention; PRO, patient-reported outcomes; RR, risk ratio; Z, overall effect size (Z-statistics); χ 2 , Cochran's Q test.

Effects on outcome measures
Many findings in this analysis corroborated with aggregated findings from previous reviews such as: using PDSIs reduces decisional conflict [7][8][9] , enhances disease and treatment knowledge [7][8][9] , and improves decision quality [8] . According to the ODSF [108] (Fig. 2), PDSIs can assist in meeting decisional needs by providing information on the possible treatment options and health conditions, as well as the associated benefits and harms [5] . This enables patients to appreciate the value-sensitive nature of decisions, thus enhancing the preference elicitation process [5] . ODSF theorizes that when adequate decisional support meets decisional needs, decision quality improves with a greater possibility of value concordance [108] . Nonetheless, a comprehensive needs assessment is required before the implementation of PDSIs in a specific patient population, as our subgroup analysis indicated that the magnitude of the benefits may vary across patient populations and PDSIs designs.
In line with previous systematic reviews, our meta-analyses demonstrated that PDSIs had no effects on PROs [7,9] . In theory, the use of PDSIs may improve PROs through two mechanisms: (1) encouraging the selection of treatment with greater PROs benefits, and (2) improving an individual's psychological well-being. The first mechanism contradicts our finding, which has shown that the use of PDSIs had no effects on treatment choice for most elective surgeries. This is because, in addition to the potential improvement in their PROs, the surgical candidate may also assess the risk of surgical adverse events. The second mechanism could work via enhanced shared decision-making [5,109] or decisions aligned with the patients' values and preferences. The insignificant effects on shared decision-making in this review contradicted the former; and the absence of trials that assessed positive psychological constructs such as satisfaction with treatment outcomes rendered the latter uncertain. Hence, future high-quality research is warranted to investigate the downstream effects of PDSIs, such as the positive psychological effects on treatment outcomes.
A recent Cochrane review [5] found that the use of PDSIs had no discernible effect on the choice of nonsurgical or invasive surgical intervention, whereas another review [7] found a marginal difference in which patients who used PDSIs were less likely to undergo surgical treatment. In line with the former review [5] , which focused exclusively on RCTs, our meta-analysis demonstrated no effect on the choice of invasive treatment. There are two plausible explanations for the contradictory result in the latter review [7] . First, the review combined experimental and observational studies, which increased the likelihood of bias. The analysis might also have overestimated the effect size and reported marginal effect due to the relatively smaller sample size (N = 2674), as compared with our review (N = 9938). Second, our review included trials with more diverse populations. PDSIs typically had a variable effect depending on the target population and the surgery being considered. Indeed, our subgroup analysis found that, while PDSIs had no influence on most elective surgery, they might be able to decrease the likelihood of some invasive procedures, most notably destination therapy LVAD placement and Figure 2. Possible mechanisms of patient decision support interventions in improving outcomes among elective surgical patients. Adapted from Ottawa Decision Support Framework. [109] lumbar spine surgery, which is consistent with previous review findings [8] . Therefore, although the overall effect of PDSIs on treatment choice was largely minimal, it varied with the type of surgery.
Consistent with another meta-analytic review [5] , our review indicated that the use of PDSIs did not incur increased use of the surgeon's consultation time. This would imply no increase in resource utilization and the likelihood of acceptance by clinicians if PDSIs were to be implemented in clinical practice. It should be noted that the timing of PDSIs administration might affect consultation time. Two included trials [83,99] , in which patients received a self-guided PDSIs a few days before the consultation, observed a shorter consultation time (g = − 0.06; 95% CI: − 0.38, 0.26). In contrast, two other trials [66,104] , in which patients received a self-guided PDSIs during the waiting time before the routine consultation, observed similar or even a longer consultation time (g = 0.10; 95% CI: − 0.16, 0.37) compared with the control group.

Effects of investigator moderators
Surprisingly, our review discovered that self-guided PDSIs appeared to be more effective than clinician-administered PDSIs in enhancing disease and treatment knowledge and possibly reducing decisional conflict. This finding could be due to several reasons. First, providing self-guided PDSIs well before the consultation allows patients more time to digest the information and prepare for discussing the decision [5] . Second, there may be a lack of clinician buy-in for clinician-administered PDSIs, resulting in less effectiveness. A qualitative study among oncological surgeons showed that although two-thirds of them were aware of PDSIs, less than half had used one during routine surgical consultations [111] . Lastly, it could be due to chance because our subgroup analysis included one trial.
Similarly, theory-guided development, value consideration, or provision of patient-reported outcome measure data showed no effects in our review. Increasingly, PDSIs are developed by taking into account the recommendations of ODSF, which postulates that high-quality decisions are typically those consistent with the patient's values [108] . A recent review stressed the need of including longitudinal PROs into the treatment decision-making process [112] . PROs are particularly relevant for patients considering elective surgeries as the main aim of the treatment is to improve functioning and well-being, or HRQoL. Given that PROs data is increasingly collected in clinical practice, incorporating such data into PDSIs becomes feasible. A possible reason for the insignificant subgroup differences of the design-related factors is that their effects were confounded or moderated by other contextual or implementation-related factors such as suboptimal protocol compliance. It is also possible that the design of PDSIs was inadequate or difficult for the user to comprehend. For example, it appears that the three PDSIs provided PROs data in the form of numerical scale scores without interpretation. Without training in psychometrics, patients are unlikely to be able to fully understand such information.

Limitations
Several limitations should be considered before interpreting these findings. First, while the comprehensive search approach lends credibility to this review, we used a broad search strategy and selected a large amount of data. Although two independent reviewers were involved, reviewer's fatigue might have led to the misclassification of records for inclusion. Second, the included trials were clinically and statistically heterogeneous, limiting their comparison. To address this issue, our study used subgroup analysis. Last, the English language restriction imposed on the RCTs might have limited the generalizability of the findings.

Implications for future research and patient decision support interventions design
In this review, most trials were classified as having some concerns due to the lack of a published protocol for assessing bias in the selection of reported outcomes. In addition, a few trials were rated as having high risk of bias for not blinding those receiving the intervention and not providing the randomization procedure. Hence, investigators assessing the efficacy of PDSIs in future trials should adhere to good trial design as well as reporting standards such as the Consolidated Standards of Reporting Trials (CONSORT) 2010 [113] .
Our review identified a significant gap in the reporting of PDSIs evaluation, including information about impartial review, citations to studies, and patient pilot testing. This made it challenging for reviewers to assess the quality of the PDSIs. Future research should develop and use a standardized International Patient Decision Aid Standards Version [26] so that the quality of PDSIs can be properly assessed. In addition, it is difficult to explore the information in the comparator group due to a lack of description. Future studies are recommended to comply with the Standards for UNiversal reporting of patient Decision Aid Evaluation (SUNDAE) checklist to ensure transparent and high-quality reports of PDSIs evaluation studies [13] .

Conclusions
This review has demonstrated that PDSIs targeting individuals considering elective surgeries had benefited their decision-making by reducing decisional conflict and increasing disease and treatment knowledge, decision-making preparedness, and decision quality. However, the quality of PDSIs varied and the certainty of evidence for many key outcomes was low. Nonetheless, these findings may be used to guide the development and evaluation of new PDSIs for use in elective surgical care. Furthermore, future high-quality research is needed to investigate the downstream treatment outcomes of PDSIs, such as the positive psychological effects of PDSIs.

Ethical approval
Not applicable as this is a review paper.

Sources of funding
None.