2.1 Patient and Public Involvement

This study did not include patient and public involvement in the selection of the topic, methods used, outcomes accessed or dissemination plans.

2.2 Literature Search Strategy

We searched MEDLINE, EMBASE, PsychINFO, CINAHL, and the Cochrane Controlled Trials Register through December 2024 with the use of a comprehensive and exhaustive search strategy developed by a health science librarian with expertise in systematic review literature searching. The details of the search strategy are available in Appendix SA.

2.3 Eligibility Criteria

Studies involving adult cancer patients being managed in healthcare centers were included. Interventions of interest were adherence to CPGs. The comparator was non-adherence to CPGs (i.e., usual care without use of adherence to guideline). The outcomes of interest included patient-related outcomes (overall survival (OS), disease-free survival (DFS), recurrence-free survival (RFS), incidence-based mortality, harm, and quality of life (QOL). We included systematic reviews (to identify primary studies), randomized controlled trials, and observational studies (cohort, before and after, and case-control). We excluded articles that were non-English articles, abstracts, conference proceedings, and studies involving guideline use for preventative or investigative procedures such as imaging or laboratory screening.

2.4 Data Screening Process

The bibliographic records from each database were uploaded into DistillerSR software [13]. All duplicate articles were removed by running the automated search in DistillerSR. The title and abstract screening were done by one reviewer (N.I.) and audited by a second reviewer (R.G.). A full-text screening of relevant studies identified from the first level of screening was done in duplicate by two reviewers (N.I. and R.G). Disagreements were resolved by discussion or with the help of a third reviewer.

2.5 Data Extraction and Risk of Bias Assessment

We extracted data on the characteristics of the included studies (e.g., last name of first author, publication year, country, study design, patient population, disease site, number of patients, intervention setting [primary, tertiary, inpatient, outpatient], guideline used, characteristic of intervention and control arms and study aim) and results on the outcomes of interest. For studies with multiple publications or similar source data, we included the publication with the longest follow up and with more complete data to avoid double counting populations or events.

The risk of bias assessment was conducted by using specific tools that are relevant to the study designs including the Cochrane risk of bias (RoB) 2 tool to assess randomized controlled trials (RCTs) [14] and the Newcastle-Ottawa Scale (NOS) for observational studies [15]. The NOS scale was used to grade studies according to a score of 0 (highest risk of bias) to 9 (lowest risk of bias). Cut-off points of < 5 (high risk of bias); 5–7 (moderate risk of bias); > 7 (low risk of bias) was applied. Data extraction and risk of bias assessment were conducted independently by two reviewers (N.I. and B.H.).

2.6 Data Analysis

Descriptive statistics, including means, frequencies and percentages, were used to summarize the main characteristics of the included studies. When appropriate, results from comparable groups of studies were combined using a random effects meta-analysis using the Review Manager software from the Cochrane Collaboration [16]. For survival outcomes (OS, RFS and DFS) comparing guideline adherent and non-adherent groups, only studies reporting hazard ratios (HR) were included in the meta-analysis. Pooled survival results are expressed as HR with 95% confidence intervals (CI) or presented narratively when statistical pooling was not feasible. HRs < 1 denotes survival advantage for guideline adherence and HRs > 1 denotes advantage for guideline non-adherence. Adjusted HRs were preferred over unadjusted estimates to minimize confounding and better reflect the independent association between adherence to clinical practice guidelines and patient outcomes. The adjustments made by the original studies commonly included covariates such as age, sex, disease stage, comorbidities, ethnicity, smoking status, socioeconomic status, and treatment modality. Statistical heterogeneity was assessed using the Cochrane Q test (with a significance threshold set at p < 0.10), I² index, and τ² tests, with the restricted maximum likelihood method used to estimate the between-study variance [17, 18]. In cases of high statistical heterogeneity (I² > 75%), subgroup analysis were performed on the pooled estimates, stratified by disease type and study duration to explore potential sources of heterogeneity. To provide an estimate of the expected range of effect sizes in future studies, a 95% prediction interval (PI) was calculated. The prediction interval incorporates both the uncertainty in the summary effect estimate and the variability across studies. The certainty of evidence (CoE) was evaluated by applying the Grading of Recommendation Assessment, Development and Evaluation (GRADE) approach [19]. A minimally contextualized approach was used to assess the certainty and the null effect was set as the threshold. A funnel plot was visually inspected to assess potential publication bias or small-study effects.

To assess the influence of individual studies on the overall meta-analytic results, a leave-one-out sensitivity analysis was conducted. In this approach, each study was sequentially removed, and the meta-analysis was re-run to observe the change in the pooled effect size, confidence intervals and heterogeneity. Further analysis was also conducted by study design by excluding the pre-post and prospective cohort studies to assess their impact on the pooled effect size.

3.1 Summary of the Included Studies

Most of the studies included patients with gastrointestinal (n = 24, 32%) and breast cancers (n = 17, 23%) and were conducted mostly in Europe (n = 38, 51%) and North America (n = 26, 35%). Eighty-eight percent of the studies were retrospective cohort studies [6, 8-10, 20-72, 83-91]. Five studies used a pre-post study design to evaluate patient outcomes before and after the introduction of CPGs [73-77]. Three prospective cohort studies evaluated the impact of CPG adherence on overall survival in patients with prostate [78], hepatocellular carcinoma [79], and pancreatic cancer [80]. One multi-center RCT focused on an adjuvant protocol for pancreatic adenocarcinoma to evaluate the impact of adherence to specified radiotherapy protocol on survival and toxicity [81]. The primary outcome reported in these studies included OS, DFS, and RFS while all studies reported data on the CPG adherence rate. There were no studies identified evaluating the impact on QOL. Thirty-two percent of the studies used the NCCN guidelines while 15% used the German National Consensus S3 Guidelines. The European Association of Urology Guidelines was about 7% while the other guidelines used ranged between 1% and 4% of the studies. A summary of the characteristics of studies included are in Table 1. Further details on individual studies are available in the Supporting Information.

3.2 Description of Studies by Disease Sites

3.3 Adverse Events

Impact of CPG adherence on toxicity was reported in a secondary analysis of an RCT on adjuvant chemotherapy and chemoradiotherapy in patients with resected pancreatic cancers which involved the use of specified radiotherapy protocol guidelines [81]. While there was no statistically significant difference between radiotherapy adherent and non-adherent arms, there was a trend toward increased toxicity for the non-adherent patients for both hematologic Grade 4 (p = 0.08) and non-hematologic toxicity (p = 0.06) in patients receiving gemcitabine.

Impact of CPG adherence on postsurgical complications were also discussed in three studies. Verbeek et al reported that adherence to the American Thyroid Association guidelines resulted in fewer cancer-related reoperations (mean = 0.24 vs. 0.60; p = 0.027) and more biochemical cure (40.9% vs. 20%; p = 0.038) in patients with medullary thyroid cancers [64]. van Vuuren et al on the other hand reported 66% of octogenarians with rectal cancer who were treated according to guidelines experienced complications vs. 34% of those treated otherwise (p = 0.02) [32]. Lastly, Foster et al evaluated adverse outcomes in non-adherence to melanoma guidelines and reported postoperative complications in 17% of patients. These complications were 3.4-fold higher for patients treated in a margin-noncompliant fashion and 2.4-fold higher for patients treated in a lymph-node-noncompliant manner (p < 0.001 for both) [72].

None of the studies reported impact of CPG adherence on incidence-based mortality or quality of life.

3.4 Study Quality Assessment

The NOS was used to assess the 74 cohort studies. The assessment showed 35% of these studies had a low risk of bias and 65% had a moderate risk of bias with no study rated as a high risk of bias. The most common sources of risk of bias were related to the selection of the nonexposed cohort, outcome assessment, and adequacy of follow-up. The RoB 2 tool was used to assess the single RCT included, and the assessment indicated a low risk of bias.

3.5 Guideline Adherence

The average CPG adherence rates by disease sites ranged between 56% and 85%. One of the studies involving the use of a breast cancer CPG reported the highest average overall CPG adherence (97%) which was defined as the number of patients who received the recommended therapy for treatment [37]. This was followed by a study in gastrointestinal (96%) [80] and gynecological cancers (90%) [59]. The lowest adherence rates were reported in studies in genitourinary (23%) [8], gynecological (26.7%) [9], and breast cancers (31%) [10].

3.6 Effect of Guideline Adherence on Overall Survival

Thirty-three studies comprising of 30 retrospective, two pre-post and one prospective studies out of the 75 studies reported data on OS with similar outcome measures (Table 2) [6, 10, 21, 22, 25, 37-42, 44-47, 50, 51, 55, 57-59, 61, 62, 66, 67, 69, 70, 73, 75, 79, 82, 84, 85]. Adherence to guidelines may improve OS based on the result from a random effects meta-analysis (HR = 0.64; 95% CI: 0.59 to 0.70; p < 0.0001, I²: 97%; low certainty of evidence(CoE)). The CoE was rated down because data was pooled from 33 observational studies encompassing diverse disease populations and varying follow up periods (Table 3 and Figure 2). The analysis revealed high statistical heterogeneity (I² = 97%, p < 0.0001) and test for subgroup differences revealed moderate heterogeneity between subgroups (I² = 46.3%), but the difference in effect sizes across subgroups was not statistically significant (p = 0.10), indicating that the grouping variable does not strongly explain the observed variation in outcomes. Furthermore, the 95% prediction interval, which accounts for both the uncertainty in the pooled effect size and the between-study heterogeneity, ranged from 0.24 to 1.04. This wide interval suggests considerable variability in effect sizes across individual studies and highlights the potential for differing results in future studies conducted under similar conditions (Figure 2).

Additional subgroup analysis was conducted using studies involving patients with a homogenous disease stage and a follow-up duration of up to 5 years. This was feasible only within the breast cancer subgroup, which had the most studies with data available for pooling (n = 14). However, only five of these studies met the criteria for inclusion in this subgroup analysis [37, 40, 41, 44, 75]. The pooled results showed that adherence to guidelines may improve OS in this more homogenous patient population HR = 0.70 (95% CI: 0.64 to 0.77); p < 0.001, I²: 0%; GRADE: low certainty of evidence) (Figure 3 and Table 3).

The leave-one-out sensitivity analysis performed showed that the exclusion of individual studies did not substantially alter the pooled effect size or reduce heterogeneity in the overall analysis, with I² values consistently ranging between 95% and 97%. This indicates that the high heterogeneity observed is not driven by any single study. However, within the disease subgroups, two studies notably influenced the pooled effect size and I² within their respective groups. In the head and neck cancer subgroup, the exclusion of Adam et al. [62] resulted in a minor change in the pooled effect size (from 0.66 to 0.69) but substantially reduced I² from 85% to 14%. Similarly, in the thoracic cancer subgroup, the exclusion of Duggan et al. [70] increased the pooled effect size from 0.79 to 0.89 and reduced I² from 80% to 0%. These results, detailed in Supporting Information Table S2 and Figure S6, suggest that the observed heterogeneity within these subgroups may be partially attributed to the influence of these individual studies.

Further subgroup analysis by study design done by excluding the pre-post and prospective cohort studies, and grouping by study duration (Supporting Information Figures S4 and S5), did not significantly alter the pooled effect size or heterogeneity, indicating that the overall findings were robust.

3.6.2 Assessment of Publication Bias

A funnel plot was created to visually assess the presence of publication bias or small-study effects. The plot appeared symmetric, suggesting that publication bias is unlikely to have influenced the results of this meta-analysis (Supporting Information S1: Figure S3).

A visual summary of key findings is presented in the infographic (Figure 4).

4.1 Strengths and Limitations

We followed a rigorous methodology according to the Cochrane Handbook of Systematic Reviews and the PRISMA checklist for reporting systematic reviews. A thorough literature search was conducted with the use of 5 electronic databases. Data extraction was detailed, and specific tools were used to evaluate the risk of bias based on study design. We also ensured that the evidence assessment was completed using GRADE to assess the certainty of evidence.

However, this study had some limitations. First, the available studies that were pooled were all retrospective registry study designs which are inherently subject to reporting and selection bias and issues with incomplete data. Second, none of the studies gathered any quality-of-life data on these patients or included any qualitative reporting on patients' or clinicians' views on guideline adherence. This could have helped get some perspective on what could be contributing to the low adherence rate in some studies.

Some other issues to note is that several of the meta-analyses showed very high I² ranging from 78% to 99%. Despite further subgroup and sensitivity analysis, the I² was still high in most instances. Some of the likely sources of heterogeneity could be the clinical (different staging within the disease sites), variations in guidelines, sample sizes, and duration of study. The reliance on a small number of studies in certain subgroup analyses, particularly those with unequal weights is also a limitation. Specifically, the subgroup analysis for the breast cancer disease site with 5 years follow up duration was based on only five studies, which limits the robustness and generalizability of the findings. Additionally, the unequal weights across these studies may have influenced the overall effect estimate, as studies with larger sample sizes or more precise estimates exert greater influence on the pooled results. While the results of the subgroup analysis are statistically significant with high heterogeneity, the small number of studies and unequal weighting warrant caution in interpreting these findings.

Furthermore, despite the overall high heterogeneity observed in the meta-analysis (I² = 97%), subgroup analyses did not fully account for all potential sources of this variation. This may be due to the variability in study design, patient population, or other unmeasured factors.

Also, about half of the studies included used either the NCCN or the German consensus guidelines. It is important to note that these guidelines are developed from consensus amongst experts rather than based on a systematic review of evidence which would offer the strongest level of evidence to support any association between adherence to these guideline recommendations and patient outcomes.

Lastly, the prediction interval (0.24 to 1.04) illustrates substantial variability in the effect sizes of included studies. This variability may reflect differences in study populations, interventions, or other factors, emphasizing the need for cautious interpretation of pooled results and tailoring of clinical decision-making to specific contexts. Issues affecting guideline adherence could also play a role in the effect sizes reported in the studies. For example, changes in guideline-recommended care over the years, variations in techniques in surgical interventions, variations in hospital protocols or procedures, and referral bias in studies done in tertiary centers could all impact the effect sizes reported. Hence the absolute effect size reported in this study should be interpreted with caution.

4.2 Research and Practice Implications

This is the first systematic review and meta-analysis of studies evaluating the impact of guideline adherence on survival in all cancer patients regardless of their disease site. This provides a knowledge database that researchers can build on in further analyzing the impact on guidelines for each disease site or to identify areas where research is lacking. The volume of research identified in this study by disease site corresponds with the global disease burden with more patients being diagnosed with breast, gastrointestinal, and gynecological cancers as opposed to head and neck or thoracic cancers. Furthermore, this systematic review serves as a resource for registry database developers or institutional record keepers to use to identify important variables that need to be captured in their databases (e.g., QOL data). The meta-analysis also shows, with a low certainty of evidence, that adherence to CPG recommendations is associated with better survival in all the disease sites with the exception of gynecological cancers where the pooled analysis did not show an OS benefit in CPG adherence. Many of the studies that could not be pooled also showed similar trends in data. Given these findings, we recommend that guideline adherence should be encouraged in clinical practice to improve patient outcomes. There is also a need for more research to be done in disease sites where this is lacking. Studies should also be of a prospective study design with predefined variables and study endpoints rather than relying on data collected retrospectively. Given the cost and time associated in developing these guidelines, findings from this study provides stakeholders and policy makers with evidence that can be used in program planning and resource allocation.