Study protocol: systematic review and meta-analysis of randomized controlled trials in first-line treatment of squamous non-small cell lung cancer

Background There is a high unmet need for effective treatments for patients with squamous non-small cell lung cancer (NSCLC). Eli Lilly and Company is conducting a phase III, randomized, multicenter, open-label study of gemcitabine plus cisplatin plus necitumumab (GC + N) versus gemcitabine plus cisplatin (GC) for the first-line treatment of patients with stage IV squamous NSCLC. Given GC is not the only treatment commonly used for the treatment of squamous NSCLC, this study was designed to compare the survival, toxicity, and quality of life outcomes of current treatment strategies for squamous NSCLC in the first-line setting. Methods/Design A systematic review and meta-analysis (including indirect comparisons) of treatments used in squamous NSCLC will be conducted to assess the clinical efficacy (overall and progression-free survival), health-related quality of life (HRQoL), and safety (grade 3–4 toxicity) of GC + N compared to other treatments used in squamous NSCLC. PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines will be followed for all aspects of this study. A systematic literature review will be conducted to identify randomized controlled trials evaluating chemotherapy treatment in first-line NSCLC. Eligible articles will be restricted to randomized controlled trials (RCTs) among chemotherapy-naïve advanced NSCLC cancer patients that report outcome data (survival, toxicity, or quality of life) for patients with squamous histology. Following data extraction and validation, data consistency and study heterogeneity will be assessed. A network meta-analysis will be conducted based on the available hazard ratios for overall and progression-free survival, odds ratios for published toxicity data, and mean difference of HRQoL scales. Sensitivity analyses will be conducted. Discussion This is a presentation of the study protocol only. Results and conclusions are pending completion of this study. Systematic review registration PROSPERO CRD42014008968


Background
Lung cancer is the leading cause of cancer-related deaths worldwide, accounting for 1.3 million deaths annually [1]. It is defined as cancer that forms in the tissues of the lung, usually in the cells lining air passages, and is divided into two main subtypes: small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC). NSCLC is the predominant subtype form and accounts for about 85% of all lung cancers [2]; it is further divided by cell histology into adenocarcinoma, squamous cell carcinoma, and large-cell carcinoma, with adenocarcinoma the currently predominant histology. Although the overall age-adjusted incidence rates for lung cancer are declining in many developed nations, lung cancer remains the leading cause of cancerrelated deaths worldwide with an overall 5-year survival rate of about 15% [3], resulting in a significant disease burden worldwide.
The treatment of lung cancer is based on the type and stage of tumor, as well as the patient's general medical condition. For patients diagnosed with early stage disease (i.e., stages I and II), surgery offers the best option for survival and cure. Adjuvant chemotherapy is increasingly used in those with stage II disease and occasionally for those with stage IB, depending on the size of the tumor. For those with stage III lung cancer, chemoradiotherapy alone or in addition to surgery is used to treat patients; however, while treatment is administered with a curative intent, the 5-year survival for patients with regional disease is approximately 26%, which decreases to 3.9% for patients with metastatic disease [3]. Treatment for patients with advanced disease tends to be palliative, although extension in survival may be achieved. The standard first-line drug treatments for advanced NSCLC, neoadjuvant, adjuvant, or chemoradiotherapy, are generally based on the combination of a second-or third-generation cytotoxic drug with a platinum agent (cisplatin or carboplatin).
There are many drug therapies available for treatment of NSCLC; however, not all current therapies are suitable for use in tumors of all histologies. The results of clinical trials have indicated that drugs such as pemetrexed have greater efficacy among patients with adenocarcinoma than those with other NSCLC histologies (e.g., squamous cell carcinoma) [4]. Other newer agents, such as bevacizumab, are indicated for adenocarcinoma because of higher toxicities observed in patients with squamous histology [5]. Drugs such as erlotinib and gefitinib are not restricted by histology, but have greater efficacy among patients with epidermal growth factor receptor (EGFR) mutations [6,7]. The frequency of EGFR mutations in patients with squamous cell carcinoma, as opposed to those with adenocarcinoma, is very low [8]. Therefore, histology-specific treatment options are limited for patients with squamous cell carcinoma, which accounts for about 25% of all non-small cell lung cancers [9].
There is thus a high unmet need for effective treatments for patients with squamous NSCLC, as disease burden is large and there is currently a lack of targeted drug therapies for NSCLC squamous cell tumors. Eli Lilly and Company is currently developing necitumumab as a first-line treatment in patients with stage IV squamous NSCLC. The current phase III study (ClinicalTrials.gov identifier: NCT00981058) is a randomized, multicenter, open-label study of gemcitabine-cisplatin chemotherapy plus necitumumab (GC + N) versus gemcitabine-cisplatin (GC) chemotherapy alone in first-line treatment of patients with stage IV squamous NSCLC. The target patient population for this trial is comprised of male and female patients with histologically or cytologically confirmed, advanced squamous NSCLC, previously untreated for metastatic disease.
The purpose of this systematic literature review and meta-analysis is to compare survival, toxicity, and quality of life outcomes of current treatment strategies with necitumumab among patients with squamous NSCLC.

Methods/Design
This systematic literature review and meta-analysis (including indirect comparisons) will be conducted of treatments used in squamous NSCLC to assess the clinical efficacy, quality of life, and safety of GC + N compared to other treatments used in squamous NSCLC. To complete this objective, the following specific aims will be pursued:

Eligibility assessment
To be eligible, published studies must meet the criteria outlined in Table 4. Briefly, eligible articles must report at least one of the following outcomes (overall survival, progression-free survival, quality of life, or toxicity) for patients with squamous NSCLC. Eligible articles must report data from randomized controlled trials published since 1995. Abstracts of all potentially eligible citations will be reviewed and excluded if it can be definitively stated that no eligibility criterion is met. All other publications will be considered potentially eligible. Full-text articles of all potentially eligible citations will be obtained and Disease terms #18 ((lung* or pulmon*) and (adenocarcinom* or squamous or "large cell" or "non-small cell")).ab,ti. 59,093 #19 Carcinoma, Non-Small-Cell Lung/ 34,861 #20 (metastatic or advanced or stage or "stage 3" or stage4 or "stage 4" or stageIII or "stage III" or StageIV or "Stage IV" or Stage?III or "Stage ?III" or Stage?IV or "Stage ?IV" or "StageIII/IV" or "Stage III/IV" or "StageIII?/IV?" or "Stage III?/IV?" or "StageIII/StageIV" or "Stage III/Stage IV" or "StageIII?/StageIV?" or "Stage III?/Stage IV?" or inoperable or in?operable or unresectable or non?resectable or "late?stage" or (metast* or advance*)).ab,ti. Rx terms #33 drug therapy/ or treatment outcome/ or ("treatment" and "response").ab,ti. or ("drug" and "efficacy").ab,ti. or outcome assessment/  reviewed to determine final eligibility. The eligibility of both the abstracts and full-text articles will be assessed independently by two reviewers using the criteria and screening matrix presented in Table 4. If the two reviewers do not agree on the eligibility of an article, a third reviewer will serve as the tie breaker. Systematic reviews and other review articles will be scanned to ensure no eligible randomized controlled trials (RCTs) are missed.

Data extraction and verification
In a process similar to that used for assessing eligibility, two reviewers will independently extract the data elements listed in Table 5 from each eligible article. These data are extensive and it is not expected, nor is it required, that all studies will report all data fields included. However, attempts to collect as extensive of data as possible will be made to increase the potential range of sensitivity and descriptive analyses. In addition to the data extraction, two reviewers will also assess bias using the Cochrane Risk of Bias Tool and will measure study quality using the Physiotherapy Evidence Database (PEDro) scale (see the "Assessment of bias and study quality" section). Data from both reviewers will be compared. If any data element does not match, the reviewers will meet and attempt to resolve the discrepancies. In cases of non-resolution, a third reviewer will be consulted. All rules and decision criteria used in the data resolution process will be recorded for quality assurance and methodological consistency purposes. To further ensure the accuracy of the extracted data, a subset of 10% of all extracted articles will be verified by an individual not involved in the data extraction process. In cases of error detection, the full database will be reviewed to ensure accuracy.

Analysis plan
A PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) diagram will be developed based on the search strategy and eligibility assessment to show the flow of included and excluded studies. The descriptive statistics from each trial of patients with squamous cell carcinoma will be included and described. These variables will include treatment group, number of patients, mean age (standard deviation), number and percent male, number and percent with stage IV disease, overall survival, progression-free survival, toxicity, and quality of life. A network diagram visually describing existing treatments for squamous NSCLC will be created after all eligible studies have been identified. However, some publications may not present data in a format that allows them to be included in the study despite otherwise meeting eligibility criteria (e.g., mixed populations not reported separately, mixed histologies not reported separately, mixed lines of therapy not reported separately). In the case of a disconnected network resulting from the absence of data for the appropriate patient population, authors of such articles will be contacted and asked to provide the needed data from their publications that would enable connection to the studied network.
The primary purpose of this study is to perform indirect and direct comparisons of GC + N versus all identified comparators for overall survival (OS) and progression-free survival (PFS). Individual hazard ratios (HR) or median time-to-event (median time) and 95% confidence intervals (90% or 99% confidence intervals will be converted to 95%) for overall survival will be included in the network meta-analysis using a Bayesian approach that ensures the preservation of randomization in the network [10]. The HR will be used as the primary unit of analyses to evaluate differences in effect size between treatment groups. Data for analysis will be extracted directly from the text of each eligible article, calculated from data included in the text, or extrapolated from the Kaplan-Meier plot according to the method of Parmar and colleagues [11]. Graphs and figures will be digitized using TechDig software and/or xyscan tool (Debian, Inc) if necessary, and digitized values will be extracted.
Heterogeneity will be explored by comparing the fixed and random effects models to ensure that the network has good properties. Additionally, heterogeneity will be explored by visual inspection of forest plots. The consistency assumption will be tested by examining network diagrams to identify any closed "loops" where inconsistencies can occur. When the network is complex with multiarm trials, the "node-splitting" approach defined by Dias and colleagues [12] will be used to identify        inconsistencies. Density plots of the posterior samples from models based on direct, indirect, and mixed evidence will be compared. In addition, the heterogeneity parameters (variance and standard deviation) and goodness of model fit measures (residual deviance and deviance information criterion (DIC), a Bayesian criterion for model comparison) between the direct and indirect models will be compared. OS and PFS data will be analyzed using a log transformation of the HR and treating this as a continuous outcome. For studies with median time information, we will also use log transformation of the median time and treat this as a continuous outcome in sensitivity analyses. HRs are preferred summary statistics to median time per Michiels and colleagues [13], and hence, the analysis will utilize HR data for the primary outcome measure.
Ideally, the literature will provide values for log (HR) and the standard error (SE) for log (HR). If the SE for log (HR) is not available, an attempt will be made to estimate the missing value from the SE for median time, assuming an exponential distribution of survival time and log (HR) = −log (median time ratio). Alternatively, an estimate of the SE for log (HR) will be made on the basis of the number of subjects with events as specified below: 1. "MedianTime" will be converted into log (median time); 2. The SE for log (median time) is estimated as (log (upper confidence limit) − log (lower confidence limit))/2/quantile (confidence level) if a treatment arm has non-missing value for all three variables; 3. If confidence limit is missing, then the number of subjects with events can be used to estimate the standard error for log (median time) as 1/sqrt(n) for a treatment arm.
Individual odds ratios and/or toxicity rates for each grade 3-4 toxicity from each study will be included, respectively, in an NMA using a Bayesian approach that ensures the preservation of randomization in the network. Odds ratios will be calculated for studies reporting toxicity rates. Prior to creating the odds ratios, we will ensure that similar versions of toxicity scaling criteria have been used. Data for analysis will be directly extracted from the text of the article or calculated from data in the text.
A network meta-analysis of GC + N to all identified comparators will be conducted for health-related quality of life (HRQoL) measures (including EQ-5D and the Lung Cancer Symptom Scale (LCSS)) during and following therapy. The most common quality of life instruments as reported across studies will be analyzed. Initial analyses will be limited to those quality of life outcomes for which GC + N data are available. For each identified measure, a standardized mean difference in quality of life outcomes from each study will be included. First, the number of trials per HRQoL instrument will be determined. If the number of trials per HRQoL instrument is 2 or more, then these data will be analyzed. For each instrument, data will be assessed according to the guidelines for that particular instrument and then pooled across studies to determine the standardized mean difference.
A meta-regression will be conducted using the key covariates of patient age and stage of disease (percent of patients with stage IV), as these variables have prognostic value in squamous NSCLC. Additional covariates may be identified following the literature review and will be considered for inclusion in post hoc analyses to control for potential bias.

Sensitivity analyses
We anticipate that some studies will not report all relevant data. In order that such studies can still be included in the analysis, we may consider imputing missing data using established methods as appropriate [14]. If imputation is made, the Bayesian model as described above will be used as the primary analysis and will be compared with analyses including the imputed values. Sensitivity analyses may be conducted to examine the effect of this method using an approach proposed by Carpenter and colleagues [15], which entails imputing missing data under a missing at random assumption, and then reweighting the imputed data to allow for non-random selection. Sensitivity analyses as outlined for OS and PFS will also be conducted for HRQoL; however, the use of disparate HRQoL instruments or assessment time points may result in an inability to evaluate the study endpoint. Sensitivity analyses will be performed to assess the robustness of the findings. At a minimum, the following analyses will be conducted if there are at least three studies available for analysis:

Assessment of bias and study quality
The risk of bias will be appraised using the Cochrane Risk of Bias Tool (www.cochrane-handbook.org). This tool was developed specifically to assess the internal validity of RCTs. It consists of the following seven criteria: 1) randomization generation, 2) allocation concealment, 3) blinding of outcome assessors, 4) blinding patients and personnel, 5) incomplete outcome data (i.e., withdrawals), 6) selective outcome reporting, and 7) other risks of bias. The final item will include fraudulent results, other methodological flaws in the RCTs, and the potential for bias. To assess publication bias, the fail-safe N will be calculated. If the number of unpublished trials that may invalidate the findings is less than five, it will be noted in the conclusions as a potential limitation of the findings. If the number of unpublished trials to invalidate the findings is five or greater, it will be noted in the results. Furthermore, funnel plot analyses will also be conducted to provide a visual representation demonstrating where unpublished data may exist. This is planned to help guide the interpretation of the study findings and the direction of bias.
Quality of selected trials for inclusion in the review will be assessed. The PEDro quality scale, an 11-item scale designed for rating the methodological quality of randomized controlled trials [16], will be used to evaluate the quality of selected trials. Here the two reviewers will independently assess studies for methodological validity prior to inclusion. Identified studies that meet the inclusion criteria will then be grouped according to the class of statin used in the trial. High quality scores will be defined as a PEDro score ≥6 and low quality scores will be defined as a PEDro score <6.
Missing data are expected in the majority of data fields collected in this meta-analysis. In cases of missing data, heterogeneity will be tested on all outcome variables to ensure that studies are comparable. Forest plots will be created for OS, PFS, toxicity, and quality of life endpoints. In the case of non-overlapping confidence intervals, the research team will discuss the need for post hoc subgroup analyses.

Discussion
The study design for this systematic review and metaanalysis is presented here to follow PRISMA standards. Industry-sponsored or industry-led studies are increasingly under scrutiny regarding transparency and risk of bias [17]. This study protocol has been designed prior to any knowledge of the study data or outcomes from existing published literature and is being disseminated in an attempt to provide the scientific community with the ability to evaluate the methods and plans of our study before it is conducted. The study protocol has been designed to meet PRISMA standards [18,19] and is being disclosed so that our methods can be retrieved and evaluated against the final analyses and interpretation of findings.
While it is almost impossible to fully anticipate the limitations of the data once they are obtained, this study has been designed in an attempt to pre-specify all primary analyses and sensitivity analyses to demonstrate the stability in results that may be discovered. However, it is possible that there will not be sufficient data to achieve all the prespecified study aims or to complete all planned analyses. There are also possible limitations in the network connections. Unlike patients diagnosed with lung cancers of nonsquamous histology, those with squamous NSCLC have not benefited from the same depth and breadth of research conducted to identify optimal treatment strategies. Therefore, via our search criteria, we are casting a wide net in the hopes of finding studies that not only investigate, but also report, outcomes for this histological subgroup.