Prognostic factors for liver, blood and kidney adverse events from glucocorticoid sparing immune-suppressing drugs in immune-mediated inflammatory diseases: a prognostic systematic review

Background Immune-suppressing drugs can cause liver, kidney or blood toxicity. Prognostic factors for these adverse-events are poorly understood. Purpose To ascertain prognostic factors associated with liver, blood or kidney adverse-events in people receiving immune-suppressing drugs. Data sources MEDLINE, Web of Science, EMBASE and the Cochrane library (01 January 1995 to 05 January 2023), and supplementary sources. Data extraction and synthesis Data were extracted by one reviewer using a modified CHARMS-PF checklist and validated by another. Two independent reviewers assessed risk of bias using Quality in Prognostic factor Studies tool and assessed the quality of evidence using a Grading of Recommendations Assessment, Development and Evaluation-informed framework. Results Fifty-six studies from 58 papers were included. High-quality evidence of the following associations was identified: elevated liver enzymes (6 studies) and folate non-supplementation (3 studies) are prognostic factors for hepatotoxicity in those treated with methotrexate; that mercaptopurine (vs azathioprine) (3 studies) was a prognostic factor for hepatotoxicity in those treated with thiopurines; that mercaptopurine (vs azathioprine) (3 studies) and poor-metaboliser status (4 studies) were prognostic factors for cytopenia in those treated with thiopurines; and that baseline elevated liver enzymes (3 studies) are a prognostic factor for hepatotoxicity in those treated with anti-tumour necrosis factors. Moderate and low quality evidence for several other demographic, lifestyle, comorbidities, baseline bloods/serologic or treatment-related prognostic factors were also identified. Limitations Studies published before 1995, those with less than 200 participants and not published in English were excluded. Heterogeneity between studies included different cut-offs for prognostic factors, use of different outcome definitions and different adjustment factors. Conclusions Prognostic factors for target-organ damage were identified which may be further investigated for their potential role in targeted (risk-stratified) monitoring. PROSPERO registration number CRD42020208049.


WHAT IS ALREADY KNOWN ON THIS TOPIC
⇒ Steroid sparing disease modifying anti-rheumatic drugs (DMARDs) are extensively used for treating inflammatory conditions, and, while effective, they can cause hepatitis, cytopenia and acute kidney injury.⇒ Three monthly monitoring blood tests are recommended to detect these adverse events early, with more frequent monitoring in those at greater risk of toxicity.Prognostic factors that may require closer monitoring are poorly understood.

WHAT THIS STUDY ADDS
⇒ This extensive systematic review ascertained prognostic factors for myelotoxicity, hepatotoxicity and nephrotoxicity due to many non-biological DMARDs and anti-tumour necrosis factor-alpha agents in a broad range of inflammatory conditions.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
⇒ Several prognostic factors for target organ damage were identified that may require more frequent monitoring when present.
RMD Open RMD Open RMD Open over 4% of adults and are usually treated with immunesuppressing drugs such as methotrexate (MTX), azathioprine (AZA) and anti-tumour necrosis factor (TNF)alpha. [1][2][3][4][5][6] Although effective, these medicines can cause drug-induced hepatitis, acute kidney injury and/ or cytopenia.][9][10] The intended purpose of blood test monitoring is to facilitate the detection of an asymptomatic adverse event, allowing treatment to be stopped before any substantial damage occurs.][10] However, these prognostic factors are either not specified 8 or mentioned anecdotally. 9 10This systematic review therefore aimed to determine which prognostic factors predict the likelihood of these specified adverse events, and thus to aid decisions on testing frequency.
The review question was: 'Which patient and treatment factors predict liver, blood and kidney related adverseevents, and related dose adjustments or discontinuations, in patients exposed to named, non-biologic and/ or biologic immune suppressing drugs for longer than 3 months?'.

METHODS
This systematic review of prognostic factor studies was conducted in accordance with PROGnosis RESearch Strategy framework (focusing on prognostic factor research) 11 and the guidance by Cochrane prognosis methods group 12 and is reported in accordance with the Preferred Reporting Items for Systematic review and Meta-Analysis guidelines. 13The inclusion and exclusion criteria using the PICOTS system (Population, Index prognostic factor, Comparator prognostic factors, Outcome, Timing, Setting) 12 are presented in supplementary material (online supplemental methods).A protocol for this review was registered with and is published in the CRD PROSPERO database.The only alteration from the published protocol was the application of the following limitations on included studies: 1995 onwards and greater than 200 participants in the entire study.

Data sources and searches
A search strategy was developed in consultation with an information specialist and the project team.Thesaurus and free-text terms for the relevant populations were combined with terms for the interventions and validated study design filters for adverse events and prognostic studies, trials, observational cohort and case-control studies. 14The following bibliographical databases were interrogated up to 31 December 2020: MEDLINE, Web of Science, EMBASE and the Cochrane library, from 1 January 1995 to 31 December 2020, without any language restrictions (online supplemental methods).Bibliographies of included studies and relevant systematic reviews were reviewed manually to identify any additional relevant studies.We excluded studies published before 1995 as inflammatory conditions were mainly treated with corticosteroids and the outcomes of patients from that era may not be relevant to the 21st century.To ensure the review was as current as possible, an update search was conducted on 5 January 2023: the same bibliographical databases were interrogated with the same search strategies but restricted to 1 January 2020 onwards.

Study selection
Three reviewers (ME, CC and JL) independently screened 10% of the sample of the titles and abstracts of citations retrieved by the original searches to compare results for accuracy and clarity of the application of the criteria.Each reviewer then screened 30% of the remaining titles and abstracts each to identify articles that satisfied the inclusion criteria and were considered for full-text screening.At the full-text screening stage, two of the three reviewers independently made a judgement on inclusion of each of the full papers (CC, ME and JL); any disagreements on inclusion were resolved by discussion and, where necessary, consultation with another reviewer (AA).For the update search, the same process was followed, but the screening was conducted by two reviewers (CC and JL), with disagreements resolved as above.

Data extraction and quality assessment
The following data were extracted based on a modified version of the [Checklist for cricial appraisal and data extraction for systematic reviews of prediction modelling studies for prognostic factors] CHARMS-PF checklist 12 : location; population and sample size; outcomes to be predicted; start and end of follow-up period; index and comparator prognostic factors; missing data; analysis; and results (estimates and corresponding SEs/SD or CI).The effect sizes of interest (eg, HRs), cut points and adjustment factors were also extracted.HRs were prioritised over rate ratios and ORs.We did not transform from one reporting scale to another.Crude (unadjusted), and estimates additionally adjusted for other patient characteristics were extracted with the latter estimates prioritised for the evidence synthesis.All data were extracted by the lead reviewers (JL and CC) and validated by at least one other reviewer (AA, TC, DvdW and MG).Any disagreements were resolved by consensus or referral to AA.No attempt was made to contact the authors of included studies to enquire about missing or incomplete data.Where estimates and 95% CI could be calculated from raw data this was calculated by one reviewer using Stata, and where only p values and group sample sizes were available the Campbell collaboration effect size calculator was used. 15he Quality in Prognostic factor Studies (QUIPS) tool 16 was used to appraise risk of bias.Judgements of high, low or unclear risk of bias for each domain were  13 For more information, visit: http://www.prisma-statement.org/RMD Open RMD Open RMD Open independently made by two reviewers (AA and JL).Any disagreements were resolved by consensus or referral to DvdW.Review findings were synthesised using an approach informed by the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework to assess the quality of the evidence (certainty in the evidence) for each prognostic factor-outcome combination. 17Evidence from randomised controlled trials (RCTs) were considered to be high quality as a starting point, and those from observational studies were considered to be low quality.The quality of evidence was then upgraded for large effect size (up one or two levels depending on the magnitude of the effect size), dose response (up one level) and downgraded one level each for high risk of bias, imprecision, inconsistency and for single study.The evidence for each outcome was assessed using this framework independently by JL and AA, with disagreements resolved through discussion.Any uncertainties were discussed with DvdW.

Data synthesis
Given the range of conditions, treatments and time points being reviewed, the studies were heterogeneous.Study characteristics and outcome data were tabulated and presented in a narrative synthesis.Also, inadequate reporting of prognostic studies, the limitations of indirect estimation methods and the uncertainties occasioned by conversion of different estimates of effect (HRs, risk ratios and ORs) indicated that the pooling of the data in a quantitative meta-analysis was not appropriate.

Small study bias
Visual assessment of potential small study bias with funnel plots was planned to be performed if effect estimates from more than 10 studies for a prognostic factor were identified for a drug and adverse event outcome pair.However, this was not found to be the case in any instance.

Role of the funding source
The sponsors were not involved in the design or conduct of the study, nor in the analysis of the data or the decision to submit the manuscript.

Patient and public involvement
Patients with personal lived experiences of inflammatory conditions were involved in prioritising the broad area of research as being of relevance to them.
Patients advised that the systematic review should include all common inflammatory conditions.
Patients and the public members have advised that the results be shared as infographics and brief video on a study website.

RESULTS
After de-duplication 16 400 titles and abstracts were reviewed.From them 2386 full-text articles were assessed for eligibility of which 54 studies reported in 56 manuscripts were eligible.Two further studies were suggested by experts.Finally, 56 studies, reported in 58 manuscripts were included in this review (figure 1).Characteristics of study populations, interventions and outcomes are presented in online supplemental tables S1 and S2.

Combinations of drug classes
4][75] One study evaluated the AZA and MTX in populations with IBD and RA, respectively 73 ; one study evaluated the use of mesalamine and infliximab and AZA in a population with IBD 74 ; and the third study 75 evaluated non-biological DMARDs or anti-TNFs in patients with PsA.Two studies were prospective cohorts with nested case-control studies, 74 75 and one study had a retrospective cohort design. 73No details of dosing regime, drug naivety or concomitant medications

Treatments Treatments Treatments
were reported for any study.The mean age of participants ranged from 39 years 74 to 51.9 years. 73

Characteristics of outcomes
In some studies outcomes were grouped under broad headings of 'myelotoxicity' or 'hepatotoxicity', while in others outcomes were more specific, for example, 'leukopenia', or 'elevation of transaminases' (online supplemental table S1).Similar outcomes were grouped together in a clinically relevant outcome group and considered together in evidence synthesis.Differences were also found in how the outcome was recorded.Most studies reported any incidence of the outcome.In some studies, discontinuation of treatment or dose adjustment associated with the outcome was reported (online supplemental material).

Quality assessment
The results of the quality assessments of all 58 included articles (56 studies) using the QUIPS (Quality In Prognosis) tool are presented in figure 2 (details in online supplemental table S3) and a summary of findings for each domain is summarised below in a narrative synthesis.

Study participation
Participants in most studies met the criteria for the specific inflammatory diseases of interest with baseline characteristics adequately described.However, where retrospective analysis of a cohort using data sets (eg, from one institution) were used, it was often unclear how many eligible participants were screened for inclusion.In these cases, it was often not possible to tell whether only participants who met the criteria for inclusion and had data for all relevant outcomes were included in the analyses, and if so, how many potentially eligible participants were excluded due to a lack of data.This presents a risk of selection bias across these studies.

Study attrition
A lack of clear reporting of attrition and missing data in the retrospective cohort studies meant that it was not possible to judge whether there was any, and if so how much data was missing for each variable of interest.This level of detail was rarely reported, leading to the risk of bias from potentially missing data.Furthermore, in these cases, it was not possible to judge whether patients whose data was missing differed in characteristics compared with those without missing data.

Prognostic factor measurement
Definitions of key prognostic factors were in the main well reported, however some studies had many potential prognostic factors (any demographic or clinical characteristic included in a multivariable analysis), and in these cases there was generally a lack of detail for how these were defined.Given the lack of reporting of missing data, any methods of imputation of missing data were rarely reported.

Outcome measurement
Most of the included studies used clear definitions of each outcome measured, with details of laboratory tests and diagnostic thresholds clearly described.However, the timing of occurrence of the outcome was mostly not reported, or only reported descriptively, precluding evidence-synthesis on the timing of the outcomes.

Adjustment for other prognostic factors
The following list of other prognostic factors (adjustment variables) was considered by the review authors to require inclusion in multivariable analyses for these studies-age, sex, body mass index (BMI), alcohol intake, concomitant immune-suppressing drugs, disease duration, chronic kidney disease and other comorbidities.While the inclusion RMD Open RMD Open RMD Open of all these variables was not an expectation, the majority (42 studies) were considered to include either no or only a selective limited choice of adjustment (existing prognostic) factors in the multivariable model.Some studies only included variables in their multivariable analysis when the unadjusted effect of those variables was significant.tatistical analysis Selective reporting of only significant results occurred in 27 studies and where this was the case, prognostic effect estimates were consequently not available for all variables leading to potential reporting bias.

STUDY FINDINGS
Several patient and treatment factors were shown to be associated strongly with an increase or decrease in the risk of liver, blood or kidney adverse events in patients taking conventional or biological immune-suppressing drugs .Evidence found to be very low quality is only reported in the supplementary materials (online supplemental tables S4a,b-21a,b).A narrative synthesis of the results prioritising at least low-quality evidence is presented here and details outlined in tables 1-3.
Tables 1-3 summarise the GRADE judgements for prognostic factors for hepatotoxicity, cytopenia and nephrotoxicity by drug type (MTX, thiopurines and anti-TNFs, respectively).The results are presented below by adverse event type.

Anti-TNF
There was high-quality evidence of baseline liver enzyme elevations [64][65][66] being associated with hepatotoxicity.There was low-quality and moderate-quality evidence from two studies of an increased risk of hepatotoxicity with increased BMI and comorbidities, respectively. 66 75There was low-quality evidence of no increased risk of hepatotoxicity with positive antinuclear antibody or rheumatoid factor (ANA or RF). 66 75rescription of other non-steroid sparing drugs were not associated with hepatotoxicity, low-quality evidence from three studies. 65 66 75iopurines There was moderate-quality evidence that pre-existing alanine transferase elevation, 70 high-quality evidence that use of mercaptopurine rather than AZA 48 58 62 increased the risk of hepatotoxicity.There was moderatequality evidence that age 48 50 59 62 and male sex 48-50 58 59 62 increased and smoking 50 58 59 did not increase the risk of hepatotoxicity.
There was low-quality evidence that disease activity, 48 58 74 concomitant biological and non-biological immune suppressing drug therapy, 50 58 59 62 AZA dose 59 62 were not associated with hepatotoxicity.
There was moderate-quality evidence from one study that included patients with IBD treated with different drugs (5-aminosalicylates, corticosteroids, AZA, anti-TNF-alpha or none) that liver steatosis was associated with hepatotoxicity. 74

Cytopenia (including neutropenia) Methotrexate
There was low-quality evidence that chronic liver disease 31 increased cytopenia risk.

Anti-TNF
There was low-quality evidence for no association between increasing age, sex or inflammatory disease type and neutropenia. 63 67There was low-quality evidence of increased risk of cytopenia from previous neutropenia, and reduced risk with increased baseline neutrophil count. 67There was moderate-quality evidence that leucopaenia or low neutrophil count at baseline were associated with increased risk of neutropenia in those prescribed conventional DMARDs +/− biologics. 23iopurines There was high-quality evidence that mercaptopurine use was associated with an increased risk of cytopenia compared with AZA. 47-49 58There was high-quality evidence that poor thiopurine metabolisers (based on thiopurine methyltransferase/nudix hydrolase (TPMT/NUDT) genotype±enzyme intermediate or low activity 46 52 53 57 ) had increased risk of cytopenia.There was low-quality evidence that female sex 46-49 57-60 was associated, and current smoking 46 47 58 59 and disease activity 47 57-59 were not associated with cytopenia.
There was moderate-quality evidence from one study that included patients with autoimmune-rheumatic disease treated with different drugs that low baseline leucopaenia and low neutrophil count were associated with neutropenia.with 7123 participants.

Low
Alcohol Evidence of increased risk with excess alcohol consumption from three studies 21 24 27 ; but no evidence of association for any alcohol intake from three studies. 20 Comorbidity composite score* Evidence of increased risk.Evidence from two studies 24 37 with 42 237 participants with aHR (95% CI) 1.12 (1.01 to 1.24) and aOR (95% CI) 1.90 (1.00 to 3.60), respectively.

Disease duration
No evidence of increased risk from three studies 22 32 36 with 925 participants.Low

Disease severity
No evidence of increased risk.Evidence from four studies 22 32 36 42 with 1214 participants.

Disease type
Evidence of increased risk with psoriasis compared with RA.Evidence from four studies 18 24 37 41 with 42 324 participants.One of the studies 18 found no evidence of association for RA compared with PsO.

Moderate
Up two for large effect size, down one inconsistency.

Elevated liver enzymes
Evidence of increased risk.Evidence from six studies 20 22 33 37 38 42 with 5931 participants.

High
Up two for large effect size.

Liver disease
Evidence of increased risk from four studies 21 22 32 33 with 5751 participants.

High
Up two for large effect size and dose response, down one inconsistency.

Serology
No evidence of increased risk.Evidence from six studies 20 22 32 36 38 42   with 2629 participants.

Inflammatory markers
No evidence of increased risk from three studies 22 32 42   with 1076 participants.Low

Folate supplementation
Evidence of reduced risk.Evidence from three studies 26 30 37 with 1551 participants.

High
Up two for large effect size.

Leflunomide
Evidence of increased risk when combined with methotrexate.Evidence from two studies 21 36 with 2242 participants.

Moderate
Up two for large effect size and dose response, down one inconsistency.

Anti-TNF
No evidence of increased risk with anti-TNF combined with methotrexate.Evidence from four studies 21 22 32 37   3550 patients.

Methotrexate dose
No evidence of increased risk.Evidence from seven studies 18

Nephrotoxicity
There was moderate-quality evidence that concomitant use of non-steroidal anti-inflammatory drugs (NSAIDs) 40 and risk factors for renal function decline 72 were prognostic for nephrotoxicity in patients prescribed MTX and anti-TNF-alpha, respectively.
Composite toxicity (treatment discontinuation with cytopenia, acute kidney injury or elevated liver enzymes) There was moderate-quality evidence that epilepsy and blood test abnormalities in the first few months of shared care prescription 34 was associated with composite toxicity in patients prescribed leflunomide and moderate-quality evidence that chronic kidney disease stage 3 35 was associated with composite toxicity in patients prescribed mycophenolate mofetil.

DISCUSSION
The review retrieved 56 studies published in 58 papers from 1995 to January 2023 that reported potential prognostic factors for common adverse events (liver, blood and kidney) in patients with a range of conditions who were prescribed immune-suppressing drugs.Most of these were designed as retrospective cohort studies.The most consistent finding was that, across drug types, baseline elevated liver enzymes were associated with increased risk of subsequent hepatotoxicity after adjusting for many other prognostic factors.The largest quantity of evidence related to prognostic factors associated with an increased risk of hepatotoxicity, with much of this from low or moderate quality evidence.The main reasons for downgrading evidence was single-study, imprecision and inconsistency.Factors shown to increase risk included BMI, age, comorbidities and the specific drug prescribed or use of concomitant drugs.These findings varied by drug type (anti-TNFs, MTX or thiopurines).Conversely, there was strong evidence that supplementation of folates was shown to reduce risk in patients prescribed MTX.Several factors were shown to predict an increased risk of cytopenia.These included previous neutropenia, comorbidities and poor metaboliser based on TPMT/NUDT genotype±enzyme intermediate or low activity.Little evidence was identified for prognostic factors for nephrotoxicity, and the quality was low, but included concomitant use of NSAIDs in those prescribed MTX.
The review was broad, with a focus on identifying risk factors for liver, blood and kidney adverse events.The strengths of this systematic review are its comprehensive inclusion of evidence that spans all relevant immunesuppressing drugs prescribed to patients with a range of conditions and up-to-date searches which retrieved evidence as recently as January 2023.Furthermore, it was conducted and reported following international guidelines by a team of highly experienced reviewers and clinicians.The limitations are that, while the search was sensitive and extensive, some relevant studies might still have been missed given the large number of therapies and populations; and despite only including studies with at least 200 participants overall, some with 25 400 participants.

Moderate
Up two for large effect size, down one inconsistency.

Smoking
No evidence of increased risk.From three studies 50 included studies still had small numbers of participants in trial arms with relevant therapies resulting in low event rates for the outcomes of interest.Studies of certain populations may have been disproportionately excluded due to generally lower sample sizes, for example, SLE.Several studies included a small number of patients with SLE alongside patients with other inflammatory conditions.Thus, while it may be possible to extrapolate the results of the review to SLE, this should be done with caution and with a low degree of certainty.Furthermore, due to the heterogeneity in the included studies, the results of the review are pooled by drug type/adverse event without disaggregating by condition.Any differences between conditions have therefore not been explored.The evidence base itself was extensive and the risk of bias was generally low or moderate according to QUIPS assessments.However, the included studies' available data and analyses for the outcomes of interest were relatively limited, with the result that quality of the evidence was assessed as low or very low according to the GRADE criteria, except for findings for prognostic factors for elevated liver enzymes.Prognostic factor findings were mainly assessed as very low quality, and this was in the main due to data being derived from small single studies, or where this was from multiple studies there was heterogeneity in outcomes, study designs, cut points used to describe prognostic factors and populations studied which also prevented meta-analysis.We did not identify any studies where prognostic factors of combination therapy were specifically addressed and the findings of this study should be extrapolated to combination therapies with caution.

CONCLUSION
Patients prescribed immune-suppressing drugs are, in general, at higher risk of liver, blood and kidney adverse events if they have a prior history of or baseline blood test abnormalities, if they have comorbidities, or if they have tested positive for indicators of poor metaboliser activity for thiopurines.Identifying patients at the earliest opportunity who are at increased risk due to these factors could potentially help to reduce the risk of adverse events and ensure blood test monitoring is appropriately adjusted.Contributors JL, CC and AA undertook the conception and design of the systematic review; acquired, analysed and interpreted the data; and drafted and approved the final manuscript.ME contributed to acquisition of data and critical revisions to the manuscript.DvdW, MG, TC and RR contributed to the methodological design and the interpretation/synthesis of data; and contributed to critical revisions to the manuscript.As submitting author, JL accepts overall guarantorship of the paper.

Figure 1
Figure 1 Preferred Reporting Items for Systematic review and Meta-Analysis (PRISMA) 2020 flow diagram (from Page et al).13For more information, visit: http://www.prisma-statement.org/

Figure 2
Figure 2 Summary of quality of included studies.Colours represent: Green -low risk of bias; yellow -moderate risk of bias; red -high risk of bias.

3 Academic
Unit of Lifespan and Population Health, School of Medicine, University of Nottingham, Nottingham, UK Acknowledgements The authors would like to acknowledge the 'Monitoring inflammatory conditions investigators', who provided valuable input on the scope of the review and interpretation of results: Professor Hywel C Williams, Professor Maarten W Taal, Professor Guruprasad P Aithal, Dr Christopher P Fox and Professor Christian D Mallen.

1
Summary of GRADE judgements: prognostic factors for hepatotoxicity, cytopenia and nephrotoxicity in those prescribed methotrexate

Table 1 Continued
Treatments Treatments Treatments

Table 2
Summary of GRADE judgements: prognostic factors for hepatotoxicity and cytopenia in those prescribed thiopurines