Web-Based Versus Usual Care and Other Formats of Decision Aids to Support Prostate Cancer Screening Decisions: Systematic Review and Meta-Analysis
- (Dental Health ???) • June 26, 2018
Background: Prostate cancer is a leading cause of cancer among men. Because screening for prostate cancer is a controversial issue, many experts in the field have defended the use of shared decision making using validated decision aids, which can be presented in different formats (eg, written, multimedia, Web). Recent studies have concluded that decision aids improve knowledge and reduce decisional conflict.
Objective: This meta-analysis aimed to investigate the impact of using Web-based decision aids to support men’s prostate cancer screening decisions in comparison with usual care and other formats of decision aids.
Methods: We searched PubMed, CINAHL, PsycINFO, and Cochrane CENTRAL databases up to November 2016. This search identified randomized controlled trials, which assessed Web-based decision aids for men making a prostate cancer screening decision and reported quality of decision-making outcomes. Two reviewers independently screened citations for inclusion criteria, extracted data, and assessed risk of bias. Using a random-effects model, meta-analyses were conducted pooling results using mean differences (MD), standardized mean differences (SMD), and relative risks (RR).
Results: Of 2406 unique citations, 7 randomized controlled trials met the inclusion criteria. For risk of bias, selective outcome reporting and participant/personnel blinding were mostly rated as unclear due to inadequate reporting. Based on seven items, two studies had high risk of bias for one item. Compared to usual care, Web-based decision aids increased knowledge (SMD 0.46; 95% CI 0.18-0.75), reduced decisional conflict (MD –7.07%; 95% CI –9.44 to –4.71), and reduced the practitioner control role in the decision-making process (RR 0.50; 95% CI 0.31-0.81). Web-based decision aids compared to printed decision aids yielded no differences in knowledge, decisional conflict, and participation in decision or screening behaviors. Compared to video decision aids, Web-based decision aids showed lower average knowledge scores (SMD –0.50; 95% CI –0.88 to –0.12) and a slight decrease in prostate-specific antigen screening (RR 1.12; 95% CI 1.01-1.25).
Conclusions: According to this analysis, Web-based decision aids performed similarly to alternative formats (ie, printed, video) for the assessed decision-quality outcomes. The low cost, readiness, availability, and anonymity of the Web can be an advantage for increasing access to decision aids that support prostate cancer screening decisions among men.
- J Med Internet Res 2018;20(6):e228
decision making; decision aid; internet; patient participation; prostate; screening
Prostate Cancer and Screening
According to the GLOBOCAN worldwide estimates of cancer incidence and mortality produced by the International Agency for Research on Cancer, there were 1,111,700 new cases of prostate cancer and 307,500 prostate cancer deaths in 2012, making prostate cancer the second most commonly diagnosed cancer in men and the fifth leading cause of cancer deaths among men .
Screening for prostate cancer remains a controversial issue, particularly after data from two major trials were released. The United States Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial found no benefits from using prostate-specific antigen (PSA) screening for prostate cancer diagnoses . The results from the European Randomised Study of Screening for Prostate Cancer with data truncated at 13 years concluded that one prostate cancer death would be avoided and 27 excess cases detected per 781 men invited for screening with PSA . Overdiagnosis was estimated to be as high as 41%. The proportion of prostate cancer that would never have led to clinical symptoms resulted in unnecessary biopsy procedures and treatment with potential side effects, which may include urinary, sexual (eg, erectile dysfunction), and gastrointestinal complications . The CAP Randomized Clinical Trial was recently published and reported no significant difference in prostate cancer mortality with PSA screening after a median follow-up of 10 years but an increase in the detection of low-risk prostate cancer cases .
Although often encouraged by media and health care providers, prostate cancer screening is currently recommended by only a few organizations. After reviewing the available evidence, the United States Preventive Services Task Force released a draft recommendation in 2017, assigning a “C” grade recommendation to prostate cancer screening in men 55-69 years old, stating that the potential benefits and adverse effects of PSA-based screening are closely balanced in that age group. The decision about whether to be screened should be an individual one based on conversations with the physician about the benefits and adverse effects of screening in order to help men make a decision based on personal values and preferences . Many experts defend a shared decision-making process involving doctor and patient, using validated decision aids. In fact, many guidelines issued by medical organizations such as the European Association of Urology , the American Cancer Society , and the American College of Physicians  support a shared decision-making process for prostate cancer screening.
Shared Decision Making and Decision Aids
According to the International Patient Decision Aids Standards Collaboration (IPDAS), decision aids are evidence-based tools designed to help people participate in decision making about health care options with the aim of improving the quality of the decision. Many study groups have focused on the development of decision aids to support shared decision making [10,11].
As established in the original IPDAS background document, two constructs are critical for establishing the effectiveness of a decision aid: (1) the quality of both the decision-making process and (2) the actual decision. For the quality of the decision-making process, five attributes are defined, all of which are measured by different scales: (1) recognizing that a decision needs to be made, (2) feeling informed about the options, (3) understanding what values matter most for the decision, (4) discussing preferences with their practitioner, and (5) being involved in decision making. Concerning the decision quality, two core attributes should be measured: (1) a patient’s knowledge of the options and outcomes and (2) agreement between the chosen option and the features that matter most for the patient .
The most recent systematic review and meta-analysis assessing the impact of decision aids for screening decisions concluded that decision aids can increase patient knowledge, make people feel clearer about their values, reduce decisional conflict, and promote an active patient role in decision making . The authors state that more studies are needed to deepen understanding of format issues such as Web-based delivery of decision aids. In addition, if new studies can be included in the systematic review, it may be possible to sort out the reasons for heterogeneity of results (eg, the format of the decision aid). Another systematic review, focusing on decision aids for prostate cancer screening, reported similar results .
Decision aids may be implemented in different formats, including written (eg, pamphlet/booklet), multimedia (eg, video, DVD), or Web-based. Syrowatka et al, in a systematic review and meta-analysis that assessed computer-based decision aids for any preference-sensitive medical decision, concluded that decision aids are associated with a significant improvement in knowledge and decrease in decisional conflict. However, results were limited by high levels of heterogeneity . Nevertheless, the scope of the latter review was broader, including any preference-sensitive medical decision. Thus, it did not specifically address prostate cancer screening. In addition, the authors included all decision aid formats that could be accessed with a computer (eg, Web-based, videobooklet, CD-ROM). With the increasing use and ease of access to the internet, the Web has been proposed as a promising way of delivering decision aids. Therefore, it is important to assess the impact of Web-based decision aids in the prostate cancer screening decision-making process, but the number of studies addressing this subject to date have been scarce and showed mixed results.
The IPDAS Collaboration identified 12 dimensions to assess quality of patient decision aids, one of which focused on the delivery of decision aids on the internet . In fact, several theories point out the potential benefits of the internet to provide broad long-term dissemination of information that can be targeted and tailored to patient needs and preferences. Hence, IPDAS emphasized that a comprehensive systematic review focusing on the internet delivery of decision aids was needed .
To our knowledge, ours is the first systematic review and meta-analysis to compare Web-based decision aids with usual care and other formats of decision aids. We sought to investigate their impact on decision quality for men making a screening decision regarding prostate cancer.
Criteria for Considering Studies for This Review
We included randomized controlled trials (RCTs) involving men who had not been previously diagnosed with prostate cancer and who were making screening decisions concerning prostate cancer. We included studies comparing Web-based decision aids to several parameters: (1) no intervention/usual care or (2) alternative decision aids formats. For interventions to be considered Web-based, they had to correspond to any program accessed over a network connection using HTTP or through a Web-based app. According to this definition, materials such as CD-ROMs or DVDs, although computer-based, were not considered Web-based. Thus, studies with such interventions were excluded. We included studies in which at least one quality of decision-making outcome (eg, knowledge, decisional conflict, and involvement in decision making) was reported. Screening behavior, either the intention to undergo PSA screening or undergoing the actual PSA screening, were defined as secondary outcomes.
Search Methods for Identification of Studies
Our search strategy for this review included searching electronic medical and social science databases: (1) PubMed, (2) Cumulative Index to Nursing and Allied Health Literature (CINAHL), (3) PsycINFO, and (4) Cochrane CENTRAL (Cochrane Central Register of Controlled Trials). Whenever possible, the search strategies (Multimedia Appendix 1) used a combination of free text and database-specific subject headings. The search was conducted in November 2016.
Searching Other Resources
We also searched trial registries (World Health Organization, National Institutes of Health, ClinicalTrials.gov), reference lists of included trials, and the Decision Aid Library Inventory.
Data Collection and Analysis
Selection of Studies
Two reviewers screened the titles and abstracts of all retrieved articles after employing the search strategy. Those included after screening were accessed in full text. Authors were contacted to clarify study eligibility. Disagreements were resolved by consensus among 3 reviewers.
Data Extraction and Management
Selection of Studies
Data extraction was performed independently by 2 reviewers. Extracted data included study design and setting, numbers, and other characteristics of study participants and interventions in addition to outcomes and other information thought to be relevant. Whenever different publications reported on the same trial, the data corresponding to the latest follow-up were included. For cluster RCTs, we collected effect estimates and standard errors from analyses that took the clustering into account. Study authors were contacted when more detailed information was needed. Disagreements were resolved by consensus. When available, data resulting from imputation were used in the analysis in accordance with an intention-to-treat approach.
Assessment of Risk of Bias in Included Studies
Assessment of risk of bias was performed using the Cochrane tool for judging risk of bias .
Measures of Treatment Effect and Data Synthesis
We used mean differences (MD) for continuous variables that were measured with the same instrument, standardized mean differences (SMD) when a similar outcome was assessed with different instruments, and relative risks (RR) for dichotomous variables. Continuous variables were standardized to a scale of 0-100. In cases where outcome data (eg, standard deviations) were missing, we tried to reach one of the study authors by email to request the complete measures. If we were unsuccessful in obtaining the data from authors, we derived standard deviations from standard errors or confidence intervals .
We analyzed studies comparing Web-based decision aids to usual care separately from studies comparing Web-based decision aids to decision aids presented in a diverse format. Review Manager 5.3.5 software was used to estimate meta-analytical-weighted treatment effects across studies . Data analysis was conducted with a random-effects model given the heterogeneity among studies being pooled.
A sensitivity analysis was conducted in order to reassess the effect measures after excluding trials classified as having high risk of bias for any of the chosen parameters and after applying the fixed-effects model. For the knowledge outcome, a sensitivity analysis was done using MDs as an alternative to SMDs. Although we opted for SMD to pool knowledge across studies since different constructs were used to measure this outcome, the use of MD could also be defensible, as the scale itself is the same.
Results of the Search and Description of Studies
The electronic database search retrieved 2406 unique citations (2536 records), and 86 additional citations were identified through other sources (Figure 1). Of the 32 full-text articles assessed, 25 were excluded (Figure 1 describes the reasons, and further details are provided in Multimedia Appendix 2). We contacted 4 investigators to clarify methodological issues and to complete the extracted data when necessary.
Figure 1. Study flow diagram. RCT: randomized controlled trial.
Seven studies met our inclusion criteria and were included in the meta-analysis. The characteristics of included studies are presented in Table 1.
The studies were published between 2003 and 2013; five were based in the United States [20-26], one in Australia , and one in the United Kingdom . A total of 4714 men with ages ranging from 45-75 years participated in the seven included studies. Five of the studies recruited men from a primary care setting [20-22,24,26], one through a radio and newspaper advertisement , and another from an industrial manufacturing worksite . Six of the studies randomized individual patients [21,22,24-26], while one used the worksite as the unit of randomization . For this review, we referred to control interventions as usual care unless they fulfill the deﬁnition of a patient decision aid. Among the included studies, five studies compared Web-based decision aid performance to the usual care [22-24,26], four studies compared Web-based decision aids to printed decision aids [20,24-26], and two studies compared Web-based decision aids to video decision aids [20,25]. In addition to containing information about the options and outcomes regarding prostate cancer screening, all decision aids used in the studies provided a values clarification tool, except for one . All included studies assessed knowledge, and six measured decisional conflict [21-26]. Others reported outcomes included anxiety [25,26], satisfaction with decision [20,24], decision-making role [21,23,25], and intention to undergo and actually undergoing PSA screening [21-26]. All studies used a parallel design, except Allen et al, who used a cluster RCT. Allen et al used a generalized estimating equations analysis and thus properly accounting for the cluster design and the possible associated unit-of-analysis errors.
Table 1. Characteristics of included studies.
Risk of Bias in Included Studies
Assessments of the risk of bias for each study are summarized in Multimedia Appendix 3 and the authors’ support for each judgment are presented in Multimedia Appendix 4. Random sequence generation was rated as being at low risk of bias in most of the studies (6/7, 86%) and unclear risk of bias in one study. Allocation concealment was considered low risk of bias in five studies (5/7, 71%) and unclear risk in the remaining two studies.
Blinding of participants and personnel was assessed as being at low risk of bias in one study (1/7, 14%), unclear risk of bias in four studies (4/7, 57%), and high risk in two studies (2/7, 29%) [20,21]. All studies were evaluated as being a low risk of bias regarding blinding of outcome assessment. All studies were rated as low risk of attrition bias that relates to incomplete outcome data. Five studies (5/7, 71%) were classified as unclear risk of bias regarding selective reporting due to the lack of information about public registration of the trial protocol. The other two studies had a registered protocol and were rated as low risk of bias for the selective reporting parameter. When assessing other sources of bias, six studies were rated as low risk of bias (6/7, 86%). The remaining study was considered unclear risk of bias as study groups were not similar in size .
Effects of Interventions
The summary of the findings is found in Multimedia Appendix 5.
All seven studies assessed patient knowledge in the meta-analysis. Studies tested knowledge through questionnaires based on the content of the decision aids. The number of correct answers was transformed into a scale ranging from 0% (no correct answers) to 100% (all correct answers).
Web-Based Decision Aids Versus Usual Care
Four studies included knowledge comparisons for this outcome in the meta-analysis. One study used a different way for grading the questionnaire (1 point for a correct answer, 0 for any unanswered item, and –1 for an incorrect answer), so data could not be transformed for the scale described above. In addition, no standard deviations could be obtained; thus data could not be pooled . Compared to the usual care, patients allocated to Web-based decision aids had higher average knowledge scores (SMD 0.46; 95% CI 0.18-0.75; Figure 2). The study that was not included in the meta-analysis showed a higher statistically signiﬁcant average score for the Web-based decision aid group in comparison with the usual care. Four of the five RCTs assessing knowledge for Web-based decision aids compared to usual care demonstrated a statistically signiﬁcant improvement in knowledge in the Web-based decision aid group [19-24]. Taylor et al reassessed knowledge at 13 months, and the Web-based decision aid group continued to register a statistically significant increase in median scores compared to the usual care group .
Web-Based Versus Printed Decision Aids
Four studies assessed knowledge for the comparison of Web-based to printed decision aids, but only data from two studies could be pooled. The scale used by Evans et al was not convertible to 0%-100% scale . Additionally, standard deviations for study results could not be obtained for the Evans et al and Krist et al studies; these studies did not find any differences between groups regarding this outcome. No differences in the average knowledge scores were found for this comparison (SMD 0.00; 95% CI –0.11 to 0.11; Figure 2) [21,26].
Figure 2. Forest plots of standardized mean differences for knowledge. A) Web-based decision aids (DA) versus usual care, B) Web-based decision aids versus printed decision aids, C) Web-based decision versus video decision aids.
Web-Based Versus Video Aids
With regard to the comparison of Web-based decision to video decision aids, the Web-based group registered lower average knowledge scores (SMD –0.50; 95% CI –0.88 to –0.12; Figure 2) for the pooled data for two studies. Frosch et al found a smaller nonstatistically significant difference between the two groups when only the participants who reviewed the complete set of materials were considered for analysis .
Six of the seven studies measured patient decisional conflict using the decisional conflict scale [20-25]. The decisional conflict scale consists of five subscales, and total scores range from 0 (no decisional conflict) to 100 (extremely high decisional conflict). When comparing Web-based aids to usual care or alternative formats of decision aids, a negative score corresponds to a reduction in decisional conflicts, which favors Web-based decision aids.
Web-Based Decision Aids Versus Usual Care
Five studies compared Web-based decision aids to usual care in terms of decisional conflict [22,23,26]. It was not possible to pool data from two studies due to lack of standard deviation of the results [21,26]. Krist et al did not find a significant difference between the two groups in contrast with the findings of Evans et al who reported a significant higher decisional conflict for the usual care group [21,26]. Frosch et al reported the results using subscales without providing standard deviation data; this study finding showed significantly higher decisional conflict for the usual care group in the subscales of “feeling informed” and “support in decision making”, and no difference was found for the subscales “uncertainty” and “having made an effective decision” . The overall MD for decisional conflict comparing Web-based decision aids versus usual care was –7.07% (95% CI –9.44 to –4.71; Figure 3).
Web-Based Decision Versus Printed Decision Aids
Four studies assessed decisional conflict by comparing Web-based decision to printed decision aids [21,24-26]. The MD for pooled data from two studies was 0.68 (95% CI –1.46 to 2.83; Figure 3). Data from the Evans et al and Krist et al studies could not be included for meta-analysis because standard deviations could not be obtained [21,26]. Reported mean scores for decisional conflicts were similar for the two groups in the Krist et al study . Evans et al did not find any statistically significant differences for decisional conflict when a Web-based decision aid was compared to a printed one .
Web-Based Decision Versus Video Decision Aids
Ilic et al did not find any statistical differences regarding mean decisional conflict scores for patients when exposed to Web-based decision compared to video decision aids .
Figure 3. Forest plots of mean differences for decisional conflict. A) Web-based decision aids (DA) versus usual care, B) Web-based decision aids versus printed decision aids.
Participation in Decision Making
Three of the seven studies evaluated participation in decision making using the Control Preferences Scale (CPS) [21,23,25], which consists of five statements (A to E), two of which reflect patient controlled decision making, another one refers to shared decision making, and the last two reflect practitioner-controlled decision making. Decision aids are intended to enhance a patient’s active role in decision making. Therefore, a pooled RR >1 for group differences in CPS statements A to C favors Web-based decision aids, and a pooled RR <1 for statements D and E also favors Web-based decision aids.
Web-Based Decision Aids Versus Usual Care
When comparing Web-based decision aids to usual care in terms of a patient-controlled or active role in the decision-making process, the pooled RR was 1.06 (95% CI 0.97-1.16; Figure 4). For the practitioner-controlled role, a pooled RR of 0.50 was obtained, which compared Web-based decision aids to usual care (95% CI 0.31-0.81; Figure 4).
Web-Based Decision Versus Printed Decision Aids
Regarding patients who assumed an active role according to the CPS, the pooled RR was 0.96 (95% CI 0.77-1.19; Figure 5). The pooled RR for the same comparison for a collaborative role in decision making was 1.12 (95% CI 0.78-1.60; Figure 5). Finally, when pooling data that compared Web-based decision to printed decision aids in terms of a passive role according to the CPS, the RR obtained was 0.83 (95% CI 0.47-1.48; Figure 5).
Web-Based Decision Versus Video Decision Aids
Ilic et al was the only study assessing participation in decisions comparing Web-based decision to video decision aids. No statistically significant differences between groups were found for active (RR 0.89; 95% CI 0.66-1.21), collaborative (RR 1.15; 95% CI 0.68-1.95), or passive patient role in decision making according to the CPS (RR 1.47; 95% CI 0.26-8.46) .