Protocol for a systematic review and meta-analysis on preoperative risk factors for failure after fixed sling implantation for post-prostatectomy stress urinary incontinence

Rational and research aim

While several systematic reviews and meta-analyses exist on the benefits and complications associated to male slings, to our knowledge no studies have systematically reviewed and meta-analysed predictors of slings failure. By systematically combining and summarizing all relevant literature, the current review and meta-analysis aim to address this research need assessing the association between preoperative risk factors and failure of fixed perineal synthetic male slings as treatment for adult male suffering from PPI, assessing the magnitude of these associations, determining the certainty in the cumulative evidence, and identifying gaps in knowledge. By helping to incorporate all available evidence into clinical practice, our results will assist healthcare professionals in patients’ selection and counselling.

Methods

Review question

The review question was defined according to the PICOTS system as proposed by the CHARMS checklist and subsequent improvement (Table 1).^11,12

Registration and reporting

The protocol for the current review has been registered on PROSPERO of the Center for the Reviews and Dissemination (CRD) (registration number: CRD42022307160). We followed the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for the development and reporting of this systematic review protocol (see Extended data for PRISMA-P checklist)^13,14 and will adhere to the PRISMA 2020 principles¹⁵ during the process of conducting and reporting this review. In performing this review we will follow the Cochrane handbook for systematic reviews of interventions,¹⁶ the recommendations of the Cochrane Prognosis Methods Group^17–19 and other guidance²⁰ for specifically conducting systematic reviews and meta-analyses of prognostic studies.

Search strategy

Systematic searches will be performed for all citations published in the following electronic databases from their inception: PubMed platform in MEDLINE, Web Of Science, SCOPUS and Cochrane CENTRAL Register of Controlled Trials. Therefore, the searches will not be limited by historical time-constraints.

Search terms and sequential strategies used in the search algorithms are displayed in Table 2. Furthermore, a scrupulous manual search will be performed in the reference lists of retrieved articles and reviews for potentially additional eligible studies.

Study selection

Applicability to the review question will be based on the following criteria:

A two-stage pilot-tested selection process will be undertaken. As a first step, relevant citations will be considered on the basis of their title and abstract. At the second step, the full-text of selected citations will be downloaded and assessed with regard to applicability and methodological quality.

Three reviewers (FG, FM, SM) will independently perform the selection process without consideration for the results. Both screening stages will be piloted on a random selection of ten citations. During both stages, disagreements will be resolved through discussion among reviewers and, if necessary, with the senior reviewer (ES) until mutual agreement is reached. All citations will be recoded in Zotero 5.0 and a Microsoft Excel spreadsheet. The selection process will be displayed in a PRISMA flow diagram.

Data extraction

Key information from each selected study will be extracted according to the CHARMS checklist¹² in its modified version suitable for reviews of prognostic factors (CHARMS-PF).²¹

Key data will be extracted in duplicate using a pre-defined data extraction form and recorded in a Microsoft Excel spreadsheet (available as Extended data: ‘Data Collection Sheet’) where the coding process for each of the variables is defined. The spreadsheet was created by two reviewers and thereafter pilot-tested and finalized on a random selection of five studies. The form will be reviewed and modified, if necessary, after group discussion with the senior reviewer.

Data items

The major groups of variables to be coded for each eligible study are as follows: (1) study’s general information (e.g., first author, journal, year of publication, country of the first author and other involved countries, study frame and design, recruitment period), (2) participant characteristics (e.g., inclusion/exclusion criteria, number of eligible and treated patients, age at baseline, body mass index, incontinence severity and aetiology, history of irradiation, bladder neck contracture), (3) intervention characteristics (e.g., type of sling, any comparator treatment), (4) outcomes (e.g., outcomes types and definitions, endpoints, event rates, patients evaluated for risk factors, attrition) and (5) risk factors and relevant quantities (evaluated variables and measurement methods, type of statistical analyses such as univariable versus multivariable, set of variables in the multivariable model, modelling method, effect size estimates with variance and p-value).

Assessment of risk of bias

The quality of studies will be assessed for risk of bias (RoB) using the Quality in Prognosis Studies (QUIPS) tool,²¹ according to the recommendations of the Cochrane Prognosis Methods Group and the suggestions provided by Grooten WJA et al.²² The QUIPS tool consists of six domains: (1) study participation, (2) study attrition, (3) prognostic factor measurement, (4) outcome measurement, (5) study confounding, and (6) statistical analysis and reporting.²¹ Each domain encompasses three to seven prompting items that are rated on a four-grade scale (yes, partial, no, unsure). Eventually, the reviewer makes an overall judgment of the RoB within each domain (high, moderate or low RoB). This tool has demonstrated satisfactory reliability.²¹

Although for QUIPS it is recommended against calculating scores for overall study quality, in systematic reviews and meta-syntheses it is recommended to report a table of included papers in which each paper is classified as having high, moderate or low RoB.²² We based this classification on the criteria proposed by Grooten et al.²²: a study satisfying low risk of bias in at least five domains and without any high risk rating will be designated as a study with an overall low RoB. If one or more domains will be classified as having high RoB, or ≥ 3 domains as moderate RoB, then this paper will be classified as high RoB. All papers in between will be classified as having moderate RoB. On the other hand, no prior criteria will be used to make the judgment of the RoB within each domain but each rater will make the decision based on a critical overall evaluation of their ratings of the included items. The QUIPS checklists (available as Extended data) will be operationalized by the reviewer team to assess and adapt the prompting items for our specific review question and assign the agreed-upon a priori relevance to each item according to recommendations on RoB assessment.^19,21,22

Two authors (FG and SM) will independently perform the quality assessment after a piloted screening of five studies. They then will meet and review their judgments for agreement. If agreement will not be reached, a third author (ES) will render the decision. Lack of adjustment is a serious bias and all studies will be judged at high RoB with regard to unadjusted effect sizes (secondary meta-analysis – see below). Overall and study-level RoB assessment will be displayed using the ROBVIS web app.²³

Meta-analysis design

Outcomes

The outcome of the present review will be the sling failure. Studies on male slings typically report two main types of treatment failures: the failure of “cure” including patients failing to achieve a complete (“dryness”) or pretty complete continence, and the failure to achieve at least an “improvement” of the baseline incontinence status (also called “social continence” or “overall success” or simply “success”). Although heterogeneity in the definitions of these outcomes is expected in sling literature, we aim to evaluate the association between risk factors and both failure to achieve cure (failure of cure - FoC) and failure to achieve overall success (failure of success – FoS). Methods of measurement will be as per included studies. “No pad use” and “use of at maximum one pad per day” will be the preferred definitions for cure and overall success, respectively, and whenever possible, to increase homogeneity, effect sizes will be calculated based on these definitions; otherwise, outcomes definitions will be as per the included studies.

Exposures

Exposures (possible risk factors) will include any preoperative variables evaluated for association with sling failure, independently from their possible causal relationships. Although our focus is mainly on “risk factors”, potential protective factors will be also included. Comparators will be absence or fewer values of all of these variables.

Effect size measures

Odds ratio (OR) with 95% confidence intervals (CI) will be reported as an overall synthesized measure of effect size. The association is reported following the convention that OR>1 indicates that the factor under evaluation is associated to an increased risk of failure, and OR<1 indicates that the factor has a protective effect. Studies on male slings are predominantly observational in nature and thus liable to confounding, thus current guidelines for prognostic studies recommend to report both crude and adjusted measures of association to better understand the role of confounding variables.¹⁸ Consequently, although most of the reported effect measures are expected to be derived from univariable (unadjusted) analyses, we will primarily focus on adjusted estimates because they presents a more accurate picture of the unique (“independent”) contribution of the risk factors to the outcome of interest.¹⁹ As a result, we will perform separate meta-analyses based on adjusted (primary meta-analysis) and unadjusted (secondary meta-analysis) ORs for both continence outcomes. Studies will be included in the primary meta-analysis only if adequate statistical control is provided to account for the effect of relevant covariates on the association between risk factors and outcomes. Adequate adjustment is defined as statistical control for at least one of two variables: pelvic irradiation and incontinence severity. These two variables are “index” risk factors usually measured in PPI literature and reported as negative prognosticators in several reviews, primary studies and guidelines on PPI surgery.^24–28 Therefore, for the other risk factors under review, it is important to understand whether they contribute additional and independent prognostic information to the index ones. For the severity of incontinence, different methods of measurement will be allowed (e.g., pad count, pad-test). The secondary meta-analysis of crude ORs will be performed separately on an expectedly much larger sample of studies, an established approach in reviews on prognostic factors to examine the robustness of results from the primary analysis, the role of comparator risk factors, and to find other risk factors not yet evaluated by multivariable analysis.^18,29–31 Meta-analyses will be performed if ORs from at least two studies could be pooled.³² Each study will contribute only one OR for metagroup and if a study reports effect sizes on several follow-ups or with different type of adjustment, we will choose based on the following criteria in order of importance: 1) sample of at least 30 patients; 2) latest follow-up; 3) the most adjusted one. Descriptive data will be used to characterize our study population. The meta-analysis design is schematically displayed in Figure 1.

Figure 1. Meta-analysis design.

Main calculations

The type of effect size measure is expected to vary between studies, as a result, harmonization will be achieved by converting all effect measures into a common metric, the OR. Effect size conversion will be performed using RevMan calculator and Psychometrica, an online tool³³ The reported standard errors will be used to calculate the inverse variance weights. When not directly reported, standard error will be calculated using confidence interval or p-value, according to Altman and Bland.³⁴ For studies reporting hazard ratios (HR), it would be incorrect to take them as an approximation of OR because the outcome (treatment failure) is not a rare event. Therefore, HRs will be converted to ORs using baseline control risk estimates from the individual studies and the methods as outlined by Symons et al.³⁵ and Shor et al.³⁶ If studies do not provide baseline risks, we will use the average risk, prevalence of the risk factor, and the relative effect to estimate the baseline risk, according to Kooter et al.³⁷ Whenever possible, in the absence of a reported effect size, we will calculate the crude OR and its standard error for dichotomous variables based on the reported or reconstructed 2 × 2 contingency table. When zero counts occurs in a cell of a 2 × 2 contingency table, the Haldane-Anscombe correction will be applied.³⁸ This is a method to avoid error in the calculation of odds ratio by adding 0.5 to all the cells of a 2 × 2 contingency table if any of the cell expectations would cause a division by zero. Continuous outcome measures such as mean difference and standardized mean difference, will be converted to OR using the Suissa’s method^39,40 or the Hasselblad and Heidges’ method,⁴⁰ respectively. Considering that continuous risk factors are sometime handled as unit (or range) of exposure and sometime are classified into distinct exposure categories, to maximize the number of studies included in each metagroup, all ORs for continuous risk factors will be converted into estimates of the OR per unit (or range) of exposure by using the Hamling’s method for linear trend calculation.⁴¹ The values (doses) assigned to each exposure category will be based on the best information available: in particular, assignments will be made according to a slightly modified parametric method proposed by Shim et al., if medians of the categories were available or computable⁴²; otherwise, a nonparametric method will be applied, using the categories midpoints.⁴³ The method used is as follows: for open smallest category, the median (or the midpoint) is set as the dose assuming that the beginning is zero (e.g., 5 if <10) or the lowest value in the study population, as appropriate; when the categories are closed by specific values, then the median (or the midpoint) is set as the dose; for open largest category, the median (or the midpoint) of the previous category minus the starting value of the previous category is added to the starting value of the last category, and the ensuing value is set as the dose. All calculations will be performed in duplicate and using Excel spreadsheets.

Statistical methods

Random-effects models

Power to detect heterogeneity is low in meta-analyses of observational studies.⁴⁴ Furthermore, inherent between-study clinical and methodological heterogeneity is expected a priori due to the characteristics of the male slings literature: different study designs (e.g., retrospective vs prospective studies), different population characteristics (e.g., aetiology of PPI) and different outcome definitions (e.g., more or less stringent definition of success and cure). As a result, in our meta-analysis it is unlikely that studies will be functionally equivalent and effect size is assumed to vary from one study to the next. Therefore, we will take a conservative approach and use random-effects models to derive the summary exposure effects and form CIs, because random-effects models account for any observed heterogeneity regardless of whether the heterogeneity is statistically significant, and allows statistical inferences to be made to a population of studies beyond those included in the meta-analysis.^45–47 The random-effects model weights the natural logarithm of each study's effect estimate by the inverse of its variance plus an estimate of the between-study variance (tau-squared) in the presence of between-study heterogeneity.^47,48 Inverse-variance weighting is a method of aggregating two or more random variables that are weighted in inverse proportion to their variance in order to minimize the variance of the weighted average. The inverse variance is roughly proportional to sample size, but is a more nuanced measure, and serves to minimize the variance of the combined effect.⁴⁹

The DerSimonian and Laird’s (DL) method of moments estimator, implemented in RevMan software, will be use to estimate the between-study variances.⁵⁰ The pooled estimates will be then represented in forest plots. The p-values will be 2-sided and a statistically significant effect will be claimed at p-value <0.05.

Because the DL approach is suboptimal and may lead to too many statistically significant results (high type I error) when the number of studies is small and there is moderate or substantial heterogeneity,⁵¹ the approach described by Hartung and Knapp (HK method) will be applied in providing the final CIs around the pooled point estimates when less than 20 studies will be available.^52,53 It has been showed that this approach consistently results in more adequate error rates (and CIs) and statistical tests, and consequently lower type I error than the DL method, more appropriately accounting for uncertainty in variance estimates, although the pooled effect estimates remain equal.^52,54 For DL model, CIs and p-values are based on the normal distribution, whereas for the HK method, they are based on the t-distribution with the degrees of freedom equal to the number of trials minus one, and a weighted version of the DL standard error.⁵² This simple and robust modification to the common random-effects meta-analysis has been recommended in current guidelines^19,55 to improve the summary results and was performed using the Excel spreadsheat provided by IntHout et al.⁵² Generally, results are more conservative with the HK method, giving wider CIs and larger p-values for the overall treatment effect, particularly in the scenario of less than 10 studies and moderate or substantial heterogeneity (I² ≥ 30%).⁵⁶

However, the HK approach is not without limitations.^56–58 When the heterogeneity is small and study sizes are imbalanced, the HK method may produce too wide, overly conservative CIs.⁵⁶ Similarly, with very few studies (<5), especially if sample sizes are quite unequal, the HK method is often over-conservative, yielding spuriously wide and uninformative CIs, thereby leading to a loss of power and increasing the chance that relevant risk factors may be missed.^{51–53,58,59} Furthermore, in case of very few studies and homogeneous study results, CIs using the HK method may be also misleadingly narrower (anti-conservative) than that of the random-effects and fixed-effect method as well.⁵⁸ As a result, we will use an adaptive, hybrid meta-analytic approach,⁵⁸ by applying both the random-effects DL and HK methods and selecting the more conservative (widest CIs) analysis to provide the final summary estimates, thus rejecting artificially narrow CIs. Furthermore, the HK method will be discarded in favour of the DL method if yielding inconclusively wide and erratic CIs. DL and fixed-effect models will be also reported together with the HK method as sensitivity analysis (see below).^51,58,59

In a random-effects meta-analysis it is also important to consider the potential effect of a treatment or an exposure within an individual study setting, as this may be different from the average effect.^60,61 This can be achieved by calculating the prediction interval (PI) that tells us how much the effect size varies in the same units as the effect size and using absolute values.⁴⁶ The term ‘prediction interval’ relates to the use of this interval to predict the possible underlying effect in a new study that is similar to the studies in the meta-analysis. A more useful interpretation of the PI is as a description of the range (dispersion) of observed effect sizes. PIs have proved a popular way of expressing the real amount of heterogeneity in a random-effects meta-analysis.⁶¹ Actually, PI is considered the most important outcome in a random-effects meta-analysis when heterogeneity is substantial, reflecting the uncertainty we expect in the summary effect if a new study is included.^60,61 The PI can be also used to estimate in a future setting the probability that a treatment or an exposure will have a true-positive or true-negative effect, and to perform better power calculations.⁶²

PIs are, however, strongly based on the assumption of a normal distribution for the effects across studies, and can be very imprecise when based on only a few studies and if these studies are small, in which case they can appear spuriously wide or spuriously narrow.^56,62 Accordingly, the use of PI is encouraged when the number of studies is reasonable, e.g. more than ten.¹⁷ Furthermore, PIs are inaccurate when there is low heterogeneity.⁵⁶ As a result, in the present meta-analysis we plan to calculate the 95% PIs when at least 10 studies are available and I² is at least 20%, using the Higgins-Thompson-Spiegelhalter equation and DerSimonian and Laird estimator of tau-squared and variance (with Hartung–Knapp variance estimator if Hartung–Knapp adjustment is applied).⁴⁶

Assessment of heterogeneity

Heterogeneity between studies will be estimated statistically by the tau-squared, the χ2-based Cochran’s Q test and the I² inconsistency Higgins’ statistic.⁶³ Statistical significance is set at the 10% level (p-value <0.1) because of low power of the Q test. I² statistic is the ratio of between-study variance over the sum of the within-study and between-study variances; it indicates the percentage of variance in observed effects that reflects variance in true effects rather than sampling error and is a recommended indicator of heterogeneity. I² statistic ranges between 0% and 100% and, in agreement with the Cochrane handbook, 0 to 40% will be interpreted as “might not be important“, 30 to 60% as “may represent moderate heterogeneity”, 50 to 90% as “may represent substantial heterogeneity and 75 to 100% as “considerable heterogeneity”.¹⁶ Clinical heterogeneity is assumed to be mainly derived from the different types of treatment (e.g., different slings and implantation techniques) and different patients characteristics (e.g., age, incontinence aetiology and severity) among included studies, while methodological heterogeneity is caused by a variety of study designs (e.g., prospective and retrospective), and diverse follow-up lengths and sample sizes. Sub-group analysis and meta-regression will be used to address the cause of heterogeneity (see below).

Analysis of subgroups or subsets

When at least 10 studies will be available on a specific risk factor, subgroup analyses and random-effects meta-regressions will be carried out, also in absence of substantial inconsistency, based on the (pre-specified) following study characteristics: sling type (Advance vs. non-Advance), baseline incontinence severity (pad use, using cut-offs of three pads per day), sample size (cut-off of 100 patients), follow-up length (cut-off of 36 months), study design (retrospective vs. prospective), outcome definition (more vs. less stringent), level of confounder adjustment (cut-off of 4 covariates) and RoB (high vs. low-moderate).^19,64 Subgroup analyses performed in the secondary meta-analysis will be useful to assess the robustness of statistically significant differences between subgroups in the primary meta-analysis.

Metabiases

To evaluate whether summary estimates are prone to small-study effects (e.g., arising from publication bias), funnel plots will be generated and Egger’s regression test will be used to evaluate the presence of asymmetry.⁶⁵ Duval and Tweedie’s nonparametric trim-fill analysis will be used to determines how many studies would need to be included in the meta-analysis to make the funnel plot symmetrical, and calculate a meta-analytic effect size that adjust for publication bias.⁶⁶ Briefly, this method initially removes the studies that are in the asymmetric part of the funnel plot, performing a new estimate of the pooled effect size. Then, using the new estimate as the axis of symmetry, studies previously removed are added again to the funnel plot, together with the same number of putative symmetric studies. A final pooled estimate is calculated based on the filled funnel plot.⁶⁶ All these analyses will be performed only on meta-analyses featuring 10 or more studies.⁶⁷

Sensitivity and influence analyses

Several sensitivity analyses will be conducted to test the robustness of the main findings. According to published recommendations, when the HK method is applied, sensitivity analyses should always be performed using the fixed-effect and/or the random-effects models to assess that HK model based estimates of CIs are actually the most conservative^51,68 Therefore, results obtained applying the HK model will be reported together with those estimated with the fixed-effect and the random-effects DL methods. Fixed-effect model result will be also reported as sensitivity analysis when only the DL method is applied, unless no heterogeneity (I²=0) was detected (in this case both methods provide the same summary estimate). As stated above, a sensitivity analysis using the fixed-effect model is particularly useful in case of very few studies where both conventional random-effects methods and other modified methods (such as the HK method) become inaccurate.^{48,51,58,59,62}

The publication bias-adjusted pooled estimates will be also used as a sensitivity analysis to assess whether the crude quantitative syntheses is robust or not (i.e. the result is reversed) to publication bias.¹⁶

A leave-one-out analysis⁶⁹ will be performed to assess the robustness of the summary estimates by removing one study at a time from meta-analyses and recalculating the pooled effect estimates to evaluate whether the meta-analysis results were biased by any individual study. The results will be considered robust if no changes in direction and statistical significance of the effect size are observed on the exclusion of any studies, and the estimates in each case are well within the CIs of the overall estimate.

A sensitivity analysis will be also performed excluding studies in the form of randomized controlled trial or published as abstract only.

Furthermore, for the assessment of potential outliers and influential studies that may influence the robustness of the findings, multiple diagnostic measures will be applied when at least three studies will be available to ensure that the conclusions do not hinge on a few unusual studies.⁷⁰ Outliers will be identified analysing the studentized deleted residuals (the deleted residual divided by its estimated standard deviation). Influential studies will be identified based on the following parameters: difference in fits value (DFFITS), which essentially indicates by how many standard deviations the predicted average effect for the ith study changes after excluding the ith study from the model fitting; Cook’s distance (D_i), a value that summarizes how much all the fitted values change when the ith study is removed; COVRATIOi value (change in the variance–covariance matrix of the parameter estimates when excluding the ith study from the model fitting (a COVRATIOi value below 1, therefore, indicates that removal of the ith study actually yields more precise estimates of the model coefficients, or equivalently, that addition of the ith study actually reduces precision); Q_E statistics.⁷⁰ If any outlier/influential studies will be identified, as further sensitivity analysis we will examine the effect of removing them from the meta-analysis.

Missing data

For missing data, study investigators will be contacted for unreported key data or additional details via electronic mail and researchgate.net if possible. Missing mean values and/or standard deviations are a common feature of meta-analyses of continuous outcome data. Calculation of missing mean values and standard deviation will be based on Wan’s methods⁷¹ or Cochrane method,¹⁶ as appropriate. Data displayed in graphs will be extracted when not retrieved otherwise using the software WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/). If necessary, imputation of the average value borrowed from one or more studies in the meta-analysis will be used for standard deviation.⁷²

Softwares

All statistical analyses will be performed using the statistical softwares Review Manager 5.4, (RevMan) [Computer program]. Version 5.4. The Cochrane Collaboration, 2020), Meta-Essentials⁷³ for metaregression and PI calculation, JASP (The JASP Team (2017). JASP (Version 0.8.1.0) [Computer software]) for publication bias and outliers/influence analyses, and Microsoft Excel for MAC (version 2016) for all other calculations.

Assessment of confidence in cumulative evidence

To interpret the confidence in cumulative evidence, we will follow the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) guidance regarding the determination of certainty in estimates of association between risk factors and outcomes.⁷⁴ This approach has been found to work well in assessing individual risk factors.⁷⁴ The quality of evidence on each risk factor and from each type of analysis (primary and secondary) will be assessed separately. A summary of finding table will be provided. Grade assessment will be conducted in duplicate with disagreements resolved through discussion and, if necessary, with the senior reviewer (ES) until mutual agreement is reached. The assessment of a body of longitudinal observational studies in the field of prognostic research begins as high certainty in the evidence.^74,75 The GRADE approach considers five factors that can decrease the confidence in estimates of effects: (1) RoB, (2) inconsistency, (3) indirectness, (4) imprecision and (5) publication bias; and three factors that can increase our confidence in estimates of effects from observational studies: (1) large estimates of treatment, (2) a dose–response gradient and (3) plausible confounding that would increase confidence in an estimate.⁷⁶ Certainty is ultimately designated as high, moderate, low, or very low. We will downgrade for RoB if ≥50% of the studies will be at high RoB or if the results are no longer significant or even reversed after removing high RoB studies. An OR higher than 2.5 will be considered as a large estimates for dichotomous or dichotomized risk factors.^77,78

Dissemination

The result of the review will be presented at national and international conferences and will be reported in peer-reviewed medical journals following the PRISMA statement.¹⁵

Study status

The research team performed a pilot search in the relevant databases in Mars 2022. This search was used to select the keywords and delineate the final search strategy, start the pilot phase of the studies selection process, customize the QUIPS tool, and create the data extraction form. The first full search and study selection has been completed. Data extraction, risk of bias assessment and data analysis are ongoing. A second full search is scheduled to be conducted in January 2023. After completion of the review, a last search will be performed before the submission of the final report.

Ethical considerations

Ethical approval is not required for this systematic review and meta-analysis because it will be based on secondary analysis of data already available in scientific databases and individual patient data will not be obtained or accessed. Even if authors of included studies will be asked to provide relevant missing data, any clinical information connecting with an individual patient will not be revealed.