A systematic review and meta-analysis of the prevalence of therapeutic targets in cervical cancer

Cervical Cancer (CC) is a significantly prevalent disease in developing countries. Currently, targeted therapies are not a primary standard of care in CC. This information could be crucial for developing directed therapies and patient screening for biomarkers that would allow personalised treatment of CC. This systematic review aimed to estimate the prevalence of potential therapeutic targets such as the epidermal growth factor receptor (EGFR) and the PI3K/Akt/mTOR and Ras/Raf/MAPK pathways in patients with CC, identified through genomic and non-genomic testing. Studies were identified through an ad-hoc search strategy from the available on MEDLINE (Ovid), CENTRAL, LILACS, SCOPUS, through the Clinical Trial registry on Clinicaltrials.gov, International Clinical Trials Registry Platform, RENIS (Argentine National Registry of Health Research) and grey literature sources. We included 74 studies which represented a total pool of 7,862 participants. Forty-five studies informed mutations of EGFR, with a combined positivity rate of 53% (95%CI: 45%–60%; I2 = 95%). Twenty studies informed the presence of mutations in PIK3CA with a combined positivity rate of 30% (95%CI: 21%–39%; I2 = 96%). Twenty-three studies reported a mutation in Ras, with a combined positivity rate of 14% (95%CI: 8%–21%; I2 = 95%). Raf mutations were informed in six studies. Six studies informed the presence of Akt mutations, two studies informed mTOR mutations and only one study reported mutations of MAPK. The most frequently described therapeutic targets were EGFR, and the PIK3CA and Ras pathways, though inconsistency in positivity rates was significant. Our study did not allow the identification of any specific clinical characteristics that might explain the observed heterogeneity. Despite the overall good quality of the included studies, the applicability of these results to patients’ general population with CC is still unclear.


Search methods
The search was performed using the following electronic databases: MEDLINE (Ovid), CENTRAL, LILACS, SCOPUS and the registry of Clinical Trials Clinicaltrials.gov, the International Clinical Trials Registry Platform (ICTRP) and RENIS until September 2019. The search strategy was designed and executed by a librarian specialised in systematic reviewing. The full search strategy, which was initially developed for MEDLINE (Ovid) and later adjusted for the other search engines, is listed in Appendix A. Searches were also performed in grey literature sources, through Google Scholar and OpenGrey. The references of the included studies were revised, and the summaries and abstracts of the most relevant speciality meetings and conferences of the past 5 years were manually explored: Annual Meeting of the American Society of Gynecologic Oncologist, Annual Meeting of the International Gynecologic Cancer Society, Annual Meeting of European Society of Medical Oncology and Annual Meeting of the American Society of Clinical Oncology.

Study selection criteria
For this systematic review, the inclusion of primary studies was considered for randomised and quasi-randomised controlled trials, observational cohort studies and cross-sectional studies, including patients of any age, diagnosed with primary, non-recurrent squamous, adenosquamous or adenocarcinoma CC, in whom expression of one or more therapeutic or potential therapeutic targets were explored. Surface protein expression and intracellular genomic expression and amplification were both considered positive results. The investigated targets were the following: EGFR and both PI3K/Akt/mTOR and Ras/Raf/MAPK pathways, with or without HPV virus typification. Studies performed in commercial cell cultures, patients with CC recurrence and those with unavailable or incomplete data to estimate the proportion of participants with a positive test were excluded. Inclusion was limited to English or Spanish reports.

Study selection procedure
Study screening and selection were performed by two independent reviewers (MGP and MFC), using the online platform COVIDENCE. The first stage involved evaluating the titles and abstracts of all records retrieved by the search strategy. The second stage consisted of the full-text revision. The full texts considered relevant after initial screening were recovered and analysed according to the research question. Reviewer discrepancies were resolved by consensus or by a third reviewer (VV).

Data extraction
Data extraction was performed by completing an ad-hoc standardised electronic form (Google Form®), after independent proofing by two reviewers (MGP and MFC). Discrepancies were resolved by consensus or by a third reviewer (VV).
The following data were collected: study design and aim, language, year of publication, inclusion and exclusion criteria and characteristics of the study population, which included sample size, age at diagnosis, tumour histology (squamous, adenosquamous or adenocarcinoma) stage of disease at diagnosis according to the International Federation of Gynecology and Obstetrics staging criteria and type of treatment. Additional variables that were documented were tumour size, parametrial invasion, depth of myocervical invasion, lymphovascular space invasion, presence of pelvic or paraaortic lymph node metastases, adjuvant treatment and HPV typification.
The primary outcome was to describe the prevalence of targets or potential targets, defined as positive expression (either genomic or nongenomic) of one or more of the following markers, independent of histologic tumour subtype or stage: EGFR and both PI3K/Akt/mTOR and Ras/Raf/MAPK pathways.

Strategies for data synthesis
Statistical analysis was performed using STATA® software (StataCorp, College Station, TX). The total percentage of variation among studies due to heterogeneity was assessed using the I² test. Low heterogeneity was defined for values between 0% and 40%, moderate for 30% through 60%, substantial between 50% and 90% and considerable heterogeneity was established between 75% and 100%. Since high heterogeneity was expected, proportions were combined through meta-analysis using a random-effects model [19]. Combined estimations were calculated with 95% confidence intervals and were represented through forest plot. We attempted to explore heterogeneity by performing prespecified subgroup analysis, for both histologic subtype and detection method (genomic versus nongenomic). However, the data were insufficient for performing these analyses or did not gather enough power to detect differences.

Sensibility analysis
Performance of sensitivity analysis was intended at the study onset, in order to explore the risk of possible selection and information bias influencing the proportion of positivity of the detection tests for the therapeutic targets. These analyses were not performed since none of the included studies was considered to be at high risk of bias in either domain.

Assessment for publication bias
Funnel plots were employed to evaluate small studies' effect whenever at least ten studies were available for a specific outcome.

Assessment of study quality and bias
The methodological quality of the included studies was independently assessed by each reviewer, by using the 'Critical Appraisal by the Joanna Briggs Institute for prevalence studies' tool, adapted ad-hoc for this review [20,21]. Through this structured and standardised instrument consisting of seven specific questions, key aspects of the study population, setting and design were explored, such as detailed population description, adequate, appropriate and representative sampling, standardised measurement of outcomes and statistical aspects, including appropriate sample size and accurate use of statistical methods for data extraction (Appendix C).

Study selection and inclusion
The study selection process is summarised in Figure 1. After executing the search strategy, which is described in Appendix A, a total of 3,935 studies were retrieved. Among these, 1,256 were automatically detected as duplicates by the reference management software and excluded. Manual search did not find additional studies.
After the initial title and abstract screening process of all the initially recovered articles, 167 studies were selected for full-text review. After this stage, 74 studies that gathered the established inclusion criteria for our revision were included. The complete list of the 93 excluded studies can be found in Appendix B. The studies are grouped according to the reason for exclusion, which included impossibility of access to the full text (in all cases, direct email contact with the author was attempted but not achieved), inappropriate study design (narrative or systematic reviews), inappropriate study population (e.g. mutations described in cell cultures or cell lines or tumour histologies other than the prespecified types) and those that did not allow confidable data extraction.
With regard to the primary outcome, 57 studies (77%) assessed the prevalence of or proportion of specific mutations, 5 (6.7%) evaluated response to treatment, 8 (10.8%) focused on patient outcome and the remaining 4 had other non-clinically applicable aims (5.4%).
English was the preferred language for publication in 71 articles (96%). The remaining 3 were in Spanish. Only 4 studies (5.4%) were multicentric, led by investigators from different countries, while 70 were performed in single countries (94.6%). The study population was grossly represented by patients with CC (64 studies; 86.4%), but four studies (5.4%) also included patients with precursor lesions and five (6.7%) involved patients with other gynaecologic cancers (including breast), and one study included other solid tumours (pancreas and colon, among others).
Thirty-seven studies (50%) detailed the type of treatment received by the study population, whether it was concomitant chemo/radiation therapy, neoadjuvant chemotherapy or primary surgery. Forty-two studies (56.7%) included data regarding patient outcome and prognosis, such as disease-free survival or overall survival. Twenty-two studies (29.7%) also stated the study population's HPV status and several specified subtypes.
In six of the included studies (8.1%), the authors declared conflicts of interest. Twenty-six declared having no disclosures whatsoever (35.1%). In the remaining studies, this information was unavailable. Regarding funding and grant support, 10.2% received public funding, 10.2% received private grants and 71.6% had a combined source, through grants from universities, laboratories or scholarships. Funding was unspecified in five studies (9.5%).

Methodological quality of the included studies
In all of the included studies, the sample population was found to adequately represent the target population (criteria 1), which means that the patients with CC included, complied with the selected subtypes' prerequisite (squamous, adenosquamous or adenocarcinoma). The majority of the studies (97%) recruited the study population adequately (criteria 2), so the risk of selection bias was considered low.
Less than 6% of the included studies reported accurate sample size calculation (criteria 3), while in most cases (94%) this parameter was not assessable, given that the method for the definition of sample size was either not informed or was not considered applicable to the primary aim of the study.
The majority of the studies (77%) accurately described the study population (criteria 4). It was partially described in 13.5% and 9.5% did not provide any information regarding the characteristics of the population. All of the included studies (100%) performed data analysis with adequate coverage of the population and used one or more of the prespecified standard methods for measuring the variable of interest (100%), which is why the risks of data loss and information bias were considered low (criteria 5 and 6).
In the vast majority of the included studies, the described statistical analysis method was adequate for the study aim (93%). Statistical analysis was unclear or uninformed in 5.4% and in one case, was considered inadequate by both reviewers (criteria 7) (Figure 2).

Revision findings
The following is a description of the results for each specific therapeutic or potential therapeutic target.

Risk of publication bias
We detected significant asymmetry in the funnel plots, which could indicate the presence of publication bias for the mutation positivity proportions of EGFR and PIK3CA, while the asymmetry was less evident for Ras (Figures 3, 6 and 10, Appendix D). Given that the plots can only be represented in outcomes that group at least ten studies, we could not assess the effect of the smaller studies for the remaining markers.

Discussion
To the authors' knowledge, this is the first systematic review evaluating the prevalence or proportion of potential and established therapeutic targets in CC. The identification of these markers may allow the development of future research with crucial clinical applicability, such as the screening for specific predictive factors at the onset of treatment, as is already being performed in other tumours. This may ultimately lead to the improvement of clinical outcomes in patients diagnosed with this disease through the use of precision medicine. One example of the benefit of using target therapy to improve outcomes in a different clinical setting is the higher progression-free survival and quality of life by targeting BRAF mutations in patients with metastatic colon cancer. This practice, which currently constitutes a standard of care, allows the identification of patients who will benefit from monoclonal antibody therapy targeting EGFR (cetuximab, panitumumab, among others) [9].
According to our findings, the most frequently explored markers in patients with CC are the EGFR and both PIK3CA and Ras pathways. We were unable to identify clinical characteristics that might justify the substantial inconsistencies observed among these studies for most markers. However, in the case of Ras, we found that the histologic subtype and the nature of the detection test employed could affect the positivity rate of this target (higher in squamous carcinoma and with the use of non-genomic testing). Because of this, the pooled estimates for the prevalence of therapeutic targets were considered to be low-certainty evidence, and the value ranges extracted from primary studies should be highlighted as the main features of this systematic review.
We considered both surface expression and intracellular gene amplification positive results since research is underway in both tumour genome and surface markers that may prove useful for the identification of clinically significant alterations that will probably become relevant in personalising patient treatment and care.
Among the limitations of our study, we can describe the limitations due to the unavailability of language translation resources, which narrowed our options to including only full texts published in English or Spanish. However, we only excluded eight studies for this reason, which is why we consider that the potential impact of this limitation is probably insignificant.
On the other hand, the methodological quality appraisal and the evaluation of the risk of bias in the collected evidence may be somewhat limited. One of the methodological difficulties of systematic reviews of prevalence resides in the diversity of instruments for the appraisal of primary studies. This is in part because there is a lack of consensus regarding which domains need to be evaluated in this study type, and there is no universally accepted assessment tool. For this review, we decided to create an ad-hoc adaptation of the ten criteria of the 'Joanna Briggs Institute Critical Appraisal Checklist for Prevalence Studies', since we considered it to be the most applicable to our study aim. This structured assessment tool was developed for any clinical condition based on the systematic review of previous instruments. Its validity, applicability and acceptability have been appreciated by experts and may be currently considered the most complete appraisal tool according to a recent review [93].
In general, the quality of the studies included in this revision was adequate. One of the greatest weaknesses that we identified through this instrument was the lack of sample size estimation, which turned out to be a frequent finding in our study pool. Most of the studies reported results in terms of proportions and had not been designed to determine the prevalence of the evaluated markers within the universe of patients with CC. Also, the primary studies did not apply sampling techniques that could guarantee the representativity of the analysed specimens, allowing the generalisation of their results.
The development of directed therapies requires the identification of adequate markers. These are specific targets or checkpoints that play a crucial role in malignant cell proliferation and growth. The assessment of the prognostic and predictive clinical value of these markers lies without the scope of this systematic review and may be an appropriate subject of further research. However, the use of these treatments in other tumours has proven a significant impact on patient outcomes and quality of life and hopefully, this may be soon applicable to CC patients as well.
From a health economics standpoint, it is important also to consider the costs associated with the systematic detection of the targets to implement precision therapies. This will require the performance of economic assessments and the development of clinical trials with patientcentred outcomes that will assess the financial burden of the specific interventions that may result beneficial in each particular context.

Conclusion
This systematic review found that the most frequently described therapeutic targets were EGFR and the PIK3CA and Ras pathways. Nonetheless, the reported estimates of positivity prevalence for these markers are inconsistent. Our study did not allow the identification of any specific clinical characteristic that might explain the observed heterogeneity, except for the Ras pathways, which showed higher rates of positivity in squamous histologic subtypes and when non-genomic assays were used for testing. Despite the overall high quality of the included studies, the applicability of these results to patients' general population with CC is still unclear.

Conflicts of interest
The authors have no conflicts of interest to declare.