A systematic review of blood biomarkers with individual participant data meta-analysis of matrix-metalloproteinase-7 in IPF

BACKGROUND
Blood derived biomarkers have been extensively described as potential prognostic markers in idiopathic pulmonary fibrosis (IPF), but studies have been limited by analyses using data-dependent thresholds, inconsistent adjustment for confounders and an array of endpoints, thus often yielding ungeneralisable results. Meta-analysis of individual participant data (IPD) is a powerful tool to overcome these limitations. Through systematic review of blood derived biomarkers, sufficient studies with measurements of Matrix Metalloproteinase-7 (MMP-7) were identified to facilitate standardised analyses of the prognostic potential of this biomarker in IPF.


METHODS
Electronic databases were searched on 12th November 2020 to identify prospective studies reporting outcomes in patients with untreated IPF, stratified according to at least one pre-specified biomarker, measured at either baseline, or change over three months. Individual participant data (IPD) was sought for studies investigating MMP-7 as a prognostic factor. The primary outcome was overall mortality according to standardised MMP-7 z-scores, with a secondary outcome of disease progression in 12 months, all adjusted for age, gender, smoking and baseline FVC.


RESULTS
IPD was available for nine studies out of twelve identified, reporting outcomes from 1664 participants. Baseline MMP-7 levels were associated with increased mortality risk (adjusted HR1.23, 95%CI 1.03;1.48, I2=64.3%) and disease progression (adjusted OR1.27, 95%CI 1.11;1.46, I2=5.9%). In limited studies, three-month change in MMP-7 was not associated with outcomes.


CONCLUSION
IPD meta-analysis demonstrated greater baseline MMP-7 levels were independently associated with an increased risk of poor outcomes in patients with untreated IPF, whilst short term changes did not reflect disease progression.


INTRODUCTION
Idiopathic pulmonary fibrosis (IPF) is a chronic progressive fibrotic lung disease of unknown aetiology that affects approximately 3 million people worldwide, with a rising incidence and a median survival from diagnosis of approximately three years. [1][2][3][4][5] Disease trajectory is variable, ranging from slow progression to rapid loss of lung function and death. 6 The most recognised biomarker of disease progression in IPF is the change in forced vital capacity (FVC) at 12 months. 7 8 However, lung function measurements have limitations, including test variability related to patient effort and confounding effects of comorbidities such as emphysema. 9 Blood derived biomarkers have been extensively described as potential prognostic markers that reflect disease severity, though none have been implemented into routine clinical practice. Studies of biomarkers have been limited by small sample sizes, inconsistent methodologies including inconsistent adjustment for confounding variables, a variety of endpoints, and analysis of outcomes using data-dependent biomarker thresholds, thus often yielding inconsistent and ungeneralisable results. 10 11 Individual patient data (IPD) meta-analyses are considered the gold standard for collecting and synthesising evidence, offering a number of advantages over traditional aggregate methods, by enabling standardisation of analyses and outcomes, consistent adjustment for potential confounding factors and robust subgroup analyses according to patient characteristics. 12 13 No published studies have utilised IPD to systematically synthesise the evidence for blood biomarkers in IPF. Through systematic review of blood derived biomarkers, sufficient studies with measurements of Matrix Metalloproteinase-7 (MMP- 7) were identified to facilitate standardised analyses of the prognostic potential of this biomarker in IPF. Thus, we explore the association between MMP-7 measured at baseline and change over three months, and clinical endpoints including mortality and disease progression in adult patients with untreated IPF.

METHODS
The systematic review was conducted in accordance with a pre-specified protocol (PROSPERO registration number: CRD42019120402) and has been reported using PRISMA-IPD (Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Individual Participant Data) guidelines. 14

Search strategy and study selection
Electronic database searches were carried out in MEDLINE (1946 to latest), Embase (1974 to latest), Google Scholar, the Cochrane Register of Controlled Trials and ClinicalTrials.gov, with the last search carried out on 12 th November 2020. Keywords and controlled vocabulary terms for "idiopathic pulmonary fibrosis" and "biomarkers", alongside search filters for prognostic studies were applied ( Figure S1). 15 Hand searches of reference lists in retrieved articles were conducted to identify further studies. Unpublished and ongoing studies were identified by searching pre-print servers including medRxiv, bioRxiv and Wellcome Open Research.
Following searches, two reviewers screened through titles and abstracts before full text review independently. Disagreements were resolved by consensus with a third reviewer.
The review included all original prospective observational studies that reported outcomes in stable or exacerbating patients aged over 18 with anti-fibrotic naïve IPF, diagnosed according to contemporaneous consensus guidelines, [16][17][18]

Data extraction and risk of bias assessment
IPD were sought from corresponding authors of studies investigating MMP-7 as a prognostic factor, using secure and encrypted electronic mail communication. A minimum of three reminders, each four weeks apart were sent. Data from sponsored clinical studies were requested through various online portals. 19-21 Requested data included participant demographics (age, gender, smoking status and baseline lung function), baseline and threemonth MMP-7 levels and outcomes including 12-month lung function and overall mortality ( Figure S2).
Where IPD were not made available, aggregate data were extracted from study publications, using a proforma and verified by a second reviewer. Data included study design, participant and biomarkers characteristics, and outcome data including sample sizes, mean values and standard deviations of biomarkers in individuals with and without the event. Time to event data were collected using adjusted hazard ratios (HR) where reported.
Risk of bias assessment was carried out independently by two reviewers using the Quality in Prognostic Studies (QUIPS) tool. 22 The QUIPS tool assesses the risk of bias across six domains: study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding and statistical analysis and reporting. All studies were included in the review irrespective of their risk of bias rating. The GRADE (Grading of Recommendations, Assessment, Development and Evaluations) framework was applied to rate the overall quality of evidence for each outcome. 23

Statistical analysis
All identified studies were included in the data synthesis, with summary tables for study characteristics. Multiple cohorts within the same study were treated as individual cohorts.
The primary outcome was overall mortality. Secondary outcomes measures included change in percent predicted FVC from baseline at 12 months and disease progression defined as 10% relative decline in FVC or death within 12 months of baseline. Hazard ratios (HR) for MMP-7 levels in predicting mortality, and odds ratios (OR) for predicting disease progression, were estimated using a two-stage IPD meta-analysis with random effects and presented as forest plots. Estimates were adjusted for a priori confounders including age, sex, smoking history, and baseline FVC. Unadjusted analyses have been presented in the supplementary material ( Figure S10). Studies with a follow up duration longer than three years were censored for survival analyses. To standardise biomarker values across studies, z scores specific to each study were calculated and analysed as exposure variables. The change in MMP-7 over threemonths was calculated where available using relative percent change from baseline.
Participants with missing data were excluded using listwise deletion. The I 2 statistic was used to evaluate statistical heterogeneity between studies. Meta-regression was conducted where sufficient studies were included to explore variability in heterogeneity according to: study design (cohort vs. randomised trial), single-centre studies, non-peer reviewed manuscripts, assay methods (ELISA vs. non-ELISA), and the type of blood samples used (serum vs. plasma).
Publication bias was assessed using funnel plot analysis and Egger's test. 24 All statistical analyses were performed using Stata 16 (Statacorp, Texas US). Due to methodological heterogeneity, marked difference in outcome measures and insufficient studies for IPD, biomarkers other than MMP-7 have been described narratively and in tables. reporting data from eleven individual cohorts and 1664 participants ( Table 1). No issues with the integrity of IPD were identified. A further 15 blood biomarkers were evaluated across the included studies, with a number of studies evaluating combinations of biomarkers (Table S1).

Searches
Risk of bias assessment of the retrieved studies identified limitations and a number of possible biases ( Figure 2, Table S2). For studies included in the MMP-7 meta-analysis, publication bias was not detected statistically, but visual inspection of funnel plots suggested publication bias was present for some of the outcomes assessed. (Figure S3 and S4). Most MMP-7 studies defined the study population specifically with clear inclusion/exclusion criteria. Biomarkers were measured consistently using the same sample matrices (plasma or serum) across included participants in each study, although details of assay platforms used to measure the analytes were frequently unreported. Outcome data were measured objectively and applied consistently to all study participants. Studies evaluating biomarkers other than MMP-7 had similar limitations and risks of bias. Blood biomarkers are known to be influenced by age and sex, as well as possible lifestyle factors such as smoking, which along with baseline lung function are all confounders upon disease outcome. 25 In approximately half of all included studies, possible confounders were not measured, and there was inconsistent adjustment in estimations where accepted confounders were measured. Moreover, in a number of studies, analyses were performed using data-dependent biomarker thresholds that were inconsistent across studies.  (Table S3). When mortality at 12 months was examined specifically, baseline MMP-7 levels were inconclusively associated with death (aHR 1.33 per SD increase, 95%CI 0.99;1.78, I 2 =59.6%) ( Figure 3B). Applying the GRADE framework, we rate the confidence in mortality estimates with moderate certainty (Table S4). Where IPD was unavailable, MMP-7 values above 5.7ng/mL were associated with increased mortality (aHR 2.18 95%CI 1.1;4.32) over a median follow up of 19 months in a study of 438 participants. 26 A further study of 57 participants found MMP-7 levels did not predict death 27 (Table S5).

Baseline blood biomarkers that predict mortality
The primary outcome of mortality was evaluated for a further 14 biomarkers in a total of 17 studies not assessed in IPD meta-analysis, with inconsistent and inconclusive findings ( Figure   6 and Table S5). Study follow up times were inconsistent, effect sizes varied with wide confidence intervals, and estimates were often unadjusted for important covariates.
Three publications from the same cohort evaluated the relationship between longitudinal biomarker measurement and mortality. [30][31][32] In both discovery and validation cohorts, a rise in CA-125 over three-months doubled the risk of death, but the remaining biomarkers were not predictive of mortality ( Figure 6 and Table S6). A validation cohort of 145 participants demonstrated replication of rising neoepitopes degraded by matrix metalloproteinases (C1M, C3M, C6M and CRPM), but the rate of change of collagen synthesis peptides was not associated with mortality. 32

Baseline biomarkers that predict disease progression and change in FVC
Ten studies measured MMP-7 levels as markers of disease progression, with eight studies totalling 1383 participants included in the IPD meta-analysis. Meta-analysis demonstrated baseline MMP-7 was associated with disease progression (aOR 1.27 per SD increase, 95%CI 1.11;1.46,I 2 =5.9%) ( Figure 4). Whilst heterogeneity was low, meta-regression identified sample assay techniques (ELISA vs. other) to be a source of heterogeneity. In subgroup analysis according to assay, the odds ratio for disease progression was estimated at 1.56 per SD increase (95%CI 1.26;1.82, I 2 =0%) when restricted to studies using ELISA ( Figure S7). When the relationship between baseline MMP-7 and relative change in FVC at 12 months was examined specifically in six studies of 891 participants, meta-analysis indicated that a 1 standard deviation greater baseline MMP-7 was associated with a -0.85% relative change in 12-month FVC percent predicted (95%CI -1.65; -0.05, I 2 =0%) ( Figure 5). We assess findings for disease progression and change in FVC outcomes with high certainty (Table S4). For studies not included in IPD meta-analysis, baseline MMP-7 values above 3.8ng/mL doubled the risk of disease progression (aHR 2.2 95%CI 1.4;3.7) over a median follow-up of 19 months in 211 participants. 33 In a further study of 57 participants, MMP-7 did not predict disease progression (Table S7).
Disease progression was evaluated for a number of other biomarkers in 19 studies that were not included in IPD meta-analysis. None were consistently predictive of disease progression, though there was significant heterogeneity in adopted definitions of disease progression, with lung function indices, mortality, transplant and acute exacerbations included in various combinations at non-unified time points ( Figure 6 and Table S7, S8).

Change in biomarkers predicting disease progression
Three studies totalling 481 participants investigating the association between MMP-7 change over three months and disease progression were included in IPD meta-analysis. Change in MMP-7 over three-months was not associated with disease progression (aOR 1.00 per percent increase, 95%CI 0.99;1.01, I 2 =22.5%) ( Figure S8), nor with change in FVC over 12 months (effect size 0.01% increase per percent MMP-7 increase 95%CI -0.07;0.08, I 2 =60.8%) ( Figure S9). In a study of 211 participants not included in IPD meta-analysis, a two-fold change in MMP-7 over four months was associated with doubling the risk of disease progression. 33 In one study, participants with progressive disease had rising concentrations of CA-125 over 3 months compared to those with stable disease, but no relationship was replicated for other biomarkers. 30 (Figure 6, Table S9)

Discussion
This systematic review of prospective studies in patients with untreated IPF identified 16 blood derived biomarkers and assessed 6 outcome variables, but there were only sufficient studies to undertake an IPD meta-analysis for MMP-7. IPD meta-analysis demonstrated baseline MMP-7 levels predicted all-cause mortality and disease progression and correlated with FVC percent predicted change over 12 months. There was a 23% greater risk of overall mortality and 27% greater risk of disease progression, per standard deviation increase in baseline MMP-7 values. An inconclusive association was observed for risk of 12-month mortality. Notably, MMP-7 levels did not seem to change longitudinally over three months, with no association observed with any of the measured outcomes. However, a study not included in quantitative synthesis suggested that in those individuals where MMP-7 does rise, there may be an associated risk in progression 33 . Mortality outcomes were rated with moderate certainty and disease progression and change in FVC outcomes with high certainty (Table S4).
Our IPD meta-analysis represents the first time it has been possible to synthesise blood biomarker findings in IPF. The meta-analysis was focused on MMP-7 as there were sufficient studies available, however individually these had yielded inconsistent results, reported datadependent thresholds and often not adjusted for confounding factors. IPD enabled analysis of MMP-7 levels as continuous variables transformed to z-scores to overcome assay variability, supported standardised definition of outcomes, and consistent adjustment for important covariates, which enabled robust and reliable conclusions. We performed twostage IPD meta-analysis, which does not assess study estimate and effects simultaneously although is considered to produce unbiased estimates, 34 and enabled modelling IPD from 1492 participants across separate secure servers and portals. Analysis of heterogeneity in IPD meta-analysis indicated that assay type was a significant contributor to heterogeneity, particularly in estimates of disease progression.
There are limitations to this review. Whilst language restrictions were not applied, two articles in Japanese were excluded as they could not be translated to English to assess inclusion criteria. We included only those studies where participants were diagnosed according to international consensus guidelines, supporting the robustness and generalisability of our findings. We excluded studies in IIPs not specific to IPF, which limits interpretation in non-IPF ILDs, although ongoing studies exploring shared mechanistic pathways will provide further insight. 35 Furthermore, by focussing on untreated IPF patients our results do not address the theranostic value of MMP-7 in relation to anti-fibrotic therapy. There was significant statistical heterogeneity in some of the outcomes, and therefore these should be interpreted with caution. We were unable to explain all the residual heterogeneity using the factors we assessed. IPD was not obtained from a limited number of suitable studies, and therefore we had to report these findings narratively.
Biomarkers of disease activity have the potential to facilitate clinical management and transform early-phase clinical trials by acting as surrogate endpoints. Dysfunctional epithelial cells contribute to fibrogenesis by secreting profibrotic mediators including matrixmetalloproteinases (MMPs), 36 responsible for degrading multiple components of extracellular matrix, activating biological mediators, and facilitating epithelial-mesenchymal transition. 37 Further research could elucidate the relationship between IPF pharmacotherapy and MMP-7, particularly to identify whether changes in MMP-7 levels may represent a biomarker of therapeutic response. From a clinical perspective, MMP-7 should be considered for implementation as a prognostic tool at the point of diagnosis, especially where lung function testing is cumbersome or unavailable.
Due to heterogeneity in study designs and reported outcomes, there were insufficient data for quantitative analysis in non-MMP-7 studies. Whilst many biomarkers showed an association with mortality in single studies, replication of effects across studies was weak. We highlight sources of considerable bias and variability. Studies were typically observational, of relatively modest size with a lack of prespecified power calculations. A number of different laboratory techniques were applied to measure biomarker levels across studies, with very few studies reporting detailed assay information, particularly with regards to measures of precision, and there was inconsistency in thresholds defining positive and negative biomarker result. Short-term changes in biomarker concentrations over three-months were often not associated with specified clinical outcomes suggesting further studies are needed before such biomarkers can be adopted clinically. Further biomarker research should focus on rigorously designed longitudinal studies with discovery and validation cohorts, using validated biomarker assays and standardised endpoints. Furthermore, it is possible that combinations of biomarkers will add granularity to our understanding of pathogenesis and prognosis of IPF and further studies evaluating their utility are needed. As further studies are published, IPD meta-analysis should be considered to produce more reliable results and support generalisability.
In summary, whilst a number of other blood biomarkers have been studied for predicting prognosis, there is currently insufficient replication to enable adoption into clinical testing, with the possible exception of MMP-7. We apply robust methodology and IPD meta-analysis to demonstrate baseline MMP-7 levels predict overall mortality and disease progression in patients with untreated IPF independent of age, gender, smoking status and lung physiology.
However, short term changes in MMP-7 over three-months offered limited prognostic value in the absence of an empirical threshold.  Original consent did not include data sharing (n = 1)    Pooled adjusted odds ratios with 95% confidence intervals for risk of disease progression, per standard deviation increase in baseline MMP-7. n denotes the number of progressors, and N represents the total number of participants included in the analysis per study. All estimates were adjusted for age, sex, smoking status, and baseline FVC. Pooled effect size with 95% confidence intervals for FVC% percent predicted relative change at 12 months, per standard deviation increase in baseline MMP-7. All estimates were adjusted for age, sex, smoking status, and baseline FVC. Figure 6 -Summary of study results. Each dot represents a study (or individual cohort in studies with more than one cohort). Green dots represent studies showing an association between the biomarker and outcome, and red dots represent studies where no association was found. Larger circles represent studies with a sample size > 100 participants, and smaller circles represent studies with sample sizes smaller than 100 participants. Outcomes where no studies were found for the listed biomarker are represented with a dash (-).          Supplementary Figure 1 -MEDLINE search strategy (last search carried out on 12 th November 2020). "OR" was used to combine search terms within each PICOS category, with "AND" used to combine search terms across PICOS categories.

Copy of email sent to authors
We would be very grateful for your assistance in undertaking a robust meta-analysis. The team at University of Nottingham (UK), led by Prof Gisli Jenkins, are conducting a systematic review and metaanalysis of blood biomarkers in IPF. The protocol for the study can be found on PROSPERO: https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=120402 As part of the review, we will conduct a meta-analysis of the association of MMP-7 levels with mortality in IPF. We have chosen this biomarker because there is sufficient published data to make it feasible and useful.
To assist with this, we would be extremely grateful if you could kindly provide us with individual patient data from your highly relevant study entitled "…" published in … We also note significant heterogeneity in disease progression definitions across individual studies, and therefore hope to meta-analyse MMP-7 level associations with a shared definition based on FVC and mortality and would also appreciate data to assist with this. We appreciate the inconvenience such requests entail, and we would like to make the process as smooth as possible, we will of course acknowledge all support.
The attached excel spreadsheet highlights the anonymised data we are seeking for each patient, where available: • MMP-7 level (baseline and 3-months) • Assay method (type of assay used) • Sample type (serum or plasma)