Types of studies

We included randomised controlled trials (RCTs) only.

Types of participants

We included participants who were prescribed drugs known to cause delayed type hypersensitivity reactions with skin involvement. We accepted participants in any setting, who were of any age, gender, and ethnicity.

Prescribed drugs included, but were not limited to: antiepileptic drugs, antiretroviral drugs, antigout drugs, and antibiotics such as beta‐lactams (penicillin, amoxicillin, piperacillin, cephalosporins) and sulphonamides (sulphamethoxazole and trimethoprim). We would have included studies that described only a subset of relevant participants. Decisions to include studies that only partially addressed the population of interest would have been documented in the review, and we would have conducted a sensitivity analysis to assess the impact of the decisions on the review's findings.

Types of interventions

We considered genetic testing for any genetic variants associated with hypersensitivity reactions using all available techniques to determine individual genotypes, and compared to standard clinical practice, which does not include a genetic test. The intervention was a randomly allocated genetic test; if the test was positive, a drug that can cause hypersensitivity was avoided. We included studies whose purpose was to genotype on the basis of likely adverse skin reaction. We excluded any participants who had previously been administered the study drugs.

Types of outcome measures

We based core outcome measures on several papers describing clinical classification of drug‐induced skin reactions, including a paper entitled 'Phenotype standardisation for immune‐mediated drug‐induced skin injury' (Pirmohamed 2011), as well as papers by the RegiSCAR (European Registry of Severe Cutaneous Adverse Reactions) consortium (Bouvresse 2012; Kardaun 2013; Sekula 2011).

We assessed clinically defined hypersensitivity reaction, immunologically confirmed hypersensitivity reactions (if skin patch testing or lymphocyte proliferation assay data were available), and long‐term sequelae (including ophthalmologic, cutaneous, or liver damage, etc.).

We have provided a full list of clinically relevant outcomes, and distinction between primary and secondary outcomes, in Appendix 1.

Electronic searches

The Cochrane Skin Information Specialist searched the following databases up to 9 July 2018 using strategies based on the draft strategy for MEDLINE in our published protocol (Alfirevic 2014):

• the Cochrane Skin Specialised Register using the search strategy in Appendix 2;

• the Cochrane Central Register of Controlled Trials (CENTRAL) 2018, Issue 6, in the Cochrane Library using the strategy in Appendix 3;

• MEDLINE via Ovid (from 1946) using the strategy in Appendix 4;

• Embase via Ovid (from 1974) using the strategy in Appendix 5; and

• LILACS (Latin American and Caribbean Health Science Information database, from 1982) using the strategy in Appendix 6.

Searching other resources

Selection of studies

Two review authors (AA and AJ) independently assessed all the titles and abstracts of publications identified by the searches to assess their eligibility. The full texts of all papers found to be eligible at this initial stage were then assessed for inclusion. Publications found to be irrelevant after reading their full text were excluded, and 'Characteristics of excluded studies' tables were prepared to identify the reasons for exclusion. Consensus on the final list of trials to include was reached by discussion.

Data extraction and management

Two review authors (AA and AJ) independently extracted data from the included studies and resolved disagreements by discussion with a third author (TC). The following information on study characteristics and methods were collected into a standardised data extraction form:

• study design;

• inclusion and exclusion criteria;

• setting;

• country;

• language of publication;

• ethnicity of participants;

• control population (comprising individuals exposed to the culprit drug with no adverse effects or individuals from a healthy population who were not exposed to medications used in the trial);

• description of genotyping techniques used;

• genotyping quality control, which included deviation from Hardy‐Weinberg equilibrium (Hardy‐Weinberg equilibrium is a crucial concept in population genetics; it predicts how gene frequencies will be inherited from generation to generation and is used as a measure of quality of genetic tests) and genotype call rate;

• age;

• gender;

• concomitant medications;

• time from exposure to the culprit drug to skin reaction;

• type of skin reaction;

• location of skin lesion;

• sequelae of adverse reactions;

• other manifestations indicating systemic involvement; and

• clinical laboratory tests.

No language translations of papers were required.

Assessment of risk of bias in included studies

Two review authors (AA and AJ) independently assessed the risk of bias in each trial. We used Cochrane's tool for assessing risk of bias (Table 8.5.a in the Cochrane Handbook for Systematic Reviews of Interventions), which is based on seven domains (Higgins 2011):

• random sequence generation (selection bias);

• allocation concealment (selection bias);

• blinding of participants and personnel (performance bias);

• blinding of outcome assessment (detection bias);

• incomplete outcome data (attrition bias);

• selective outcome reporting (reporting bias); and

• other risk of bias.

The tool allows for the risk of bias for each domain to be assessed as 'low', 'high', or 'unclear' (indicating lack of information or uncertainty over the potential for bias). An additional review author (MP) was consulted in the case of disagreements.

Measures of treatment effect

We planned to use statistical methods in accordance with theCochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We planned to calculate mean difference (MD) with 95% confidence interval (CI) for continuous data or indeed standardised mean difference (SMD) with 95% CI where all studies reported an outcome using different but similar scales. A risk ratio (RR) with 95% CI was calculated for dichotomous data.

Unit of analysis issues

There were no unit of analysis issues. If such issues arise in an update of this review, we plan to take the appropriate approach to analysis to avoid a unit of analysis error due to repeated observations on participants, multiple treatments, or re‐occurring events. In future updates, if we include cluster‐randomised trials, we will take into account issues such as recruitment bias, baseline comparability of clusters, and number of clusters, and make sure that appropriate statistical methods are used that take into account weighting, etc. (Higgins 2011).

Dealing with missing data

We considered the possible different types of missing data. We planned to deal with missing studies (and the associated risk of bias) by assessing for publication bias, and to deal with missing outcomes (and the associated risk of bias) by assessing for selective reporting (see Assessment of reporting biases). However this was not applicable, due to there only being one included study in this review.

The included study did not report on either of our primary outcomes, but instead reported on an outcome of hypersensitivity reaction which included (but was not limited to) our secondary outcomes HSS and SJS/TEN. So in order to deal with missing data we contacted the study author for additional data on the outcomes of our review. We did not receive a response and therefore we are unaware if the outcomes were measured, and were unable to include these data in our review.

Assessment of heterogeneity

Had a meta‐analysis been possible, we planned to assess the extent of heterogeneity using the I² statistic. We would have used the following thresholds for the interpretation of the I² statistic:

• 0% to 40% = might not be important;

• 30% to 60% = moderate heterogeneity;

• 50% to 90% = may represent substantial heterogeneity; and

• 75% to 100% = considerable heterogeneity.

Due to there being only one included study, we did not need to assess heterogeneity. If eligible studies become available in the future during an update of this review, we will perform a meta‐analysis, report the I² statistic, and interpret the two data together.

Assessment of reporting biases

We planned to assess publication bias by using a funnel plot, Begg test and Egger test (Begg 1989; Egger 1998). Tests for funnel plot asymmetry would only be undertaken when at least 10 studies could be included in the meta‐analysis. Due to the fact that there was only one included study, the power of the tests is too low to distinguish chance from real asymmetry; therefore, this was not assessed.

Data synthesis

Had there been sufficient studies and no significant clinical heterogeneity, we planned to synthesise the results in meta‐analyses, stratified according to type of intervention (e.g. type of genetic test). A random‐effects model would have been assumed.

Where events are rare, a random‐effects approach may be inappropriate. Where events were rare, we planned to take extra care to adopt appropriate methods of meta‐analysis, as recommended in the Cochrane Handbook for Systematic Reviews of Interventions (section 16.9) (Higgins 2011), since many meta‐analysis methods are suboptimal where events are rare through results being biased, confidence intervals being too wide, or power being too low. Choice of method would have been guided by control group risk, likely treatment effect size, and consideration of balance in numbers of treated and control participants in the constituent studies. Had the control groups differed — e.g. if they were drawn from a healthy population or from a population of people treated with the culprit drug but without any adverse effects — we planned to conduct separate analyses. We planned to use Review Manager 5 software to undertake the meta‐analyses (Review Manager 2014).

As we only included one study, it was not possible to undertake a meta‐analysis.

Subgroup analysis and investigation of heterogeneity

We planned to assess heterogeneity using the I² statistic and in the event of substantial heterogeneity, we planned to explore the causes by way of subgroup analyses. We planned to consider the following:

• participant factors (age, ethnicity, classification of adverse drug reactions, and comparability of participant groups); and

• trial design issues (genotyping methodology and quality control, blinding, drugs included, and drug dosage and duration of use).

As we only included one study, it was not possible to assess for heterogeneity and therefore a subgroup analysis was not relevant.

Sensitivity analysis

We planned to evaluate the robustness of the results of the meta‐analyses by removing trials of low methodological quality as identified by their risk of bias. As we only included one study, it was not possible to undertake a meta‐analysis, or indeed sensitivity analyses.