Assessing similarity of studies for network meta-analysis: the use of feasibility assessment and sensitivity analyses to estimate the efficacy of tirbanibulin, a new drug for actinic keratosis

We conducted a systematic literature review and network meta-analysis (NMA) to assess complete clearance of actinic keratosis by a new drug, tirbanibulin 1% ointment, compared with existing treatments in Europe. A feasibility assessment (FA) was conducted to assess the similarity of populations, interventions, comparators, outcomes and design among the 40 eligible randomized controlled trials (RCTs) relevant for European approved posologies. Five RCTs were deemed unsuitable for the NMA following FA. In the 35 remaining trials, potential treatment-effect modifiers were: number of treatment cycles, treatment area size, timepoint of response assessment, and type of placebo/vehicle. Sensitivity analyses were conducted to assess the impact of these variables. In the primary analysis, all odds ratios credible intervals overlapped, suggesting similar efficacy across treatments (tirbanibulin, imiquimod 5%, imiquimod 3.75%, fluorouracil 0.5% + salicylic acid, diclofenac sodium 3%, fluorouracil 5%, fluorouracil 4%, ingenol mebutate 0.015%, cryosurgery, photodynamic therapy with aminolevulinic acid and with methyl aminolevulinate). All treatments had higher odds than topical placebo/vehicle. When analysing studies assessing a single course of treatment or treatment area of ≤25cm2 or when pooling topical and photodynamic therapy placebo nodes we obtained results consistent with the primary analysis. When including studies assessing clearance ≥8 weeks from the end of treatment, the only impact was for imiquimod 5%: decreased performance, possibly due to decreased size of the evidence base. Assessment of inconsistency showed that indirect posterior odds ratios were generally consistent with direct posterior odds. Results of the primary analysis suggested that tirbanibulin can achieve similar complete clearance to other treatments for AK. The treatment-effect modifiers were found to have little impact on the results, increasing confidence in their generalizability within the population of interest. Identifying potential treatment-effect modifiers and quantitatively assessing their impact on the results helps to draw more robust conclusions from NMAs.