2.5 Data extraction and management
Two independent review authors (PA & SM) screened the titles and abstracts of identified studies and full texts of potentially eligible studies. Disagreements were resolved by discussion or arbitration by a third author (AM). Duplicates were removed both automatically using Endnote and by manual screening.
The two review authors independently extracted data from eligible studies into a piloted, standardised data extraction form (Microsoft Excel). If consensus was not reached, disagreements were resolved by discussion first and then arbitration by a third author (AM). Data extraction included bibliometric data (e.g. authors, title, country), study characteristics (e.g. setting, sample size, target population, funding, conflicts of interest), participant characteristics (e.g. age, sex, diagnosis, baseline number of gabapentinoids prescribed), interventions and controls (gabapentinoid type, dose, duration, intervention design and aim), co-interventions (e.g. use of other therapies), outcome data (pre/post gabapentinoid prescribing rates, adverse events including withdrawal symptoms (with descriptors and related characteristics), withdrawals from the intervention, patient-reported outcomes (pre/post intervention pain intensity (e.g. Visual Analogue Scale) and quality of life (e.g. Hamilton Anxiety Rating Scale, EuroQol-5-Dimension scores), barriers to deprescribing (for patients and clinicians) and data completeness (i.e. percentage of missing data, how missing data were handled).
Serious adverse events were defined as events that were life threatening, such as those that resulted in death, hospitalisation, significant incapacity, congenital anomaly, or birth defects. Adverse events were defined as non-serious adverse events, such as side effects e.g. dizziness, drowsiness, confusion, dry mouth. Successful cessation of gabapentinoid therapy was defined as ceasing the medication without switching to another high-risk medication from another inappropriate drug class (e.g. benzodiazepine or opioid).
Follow-up time points for outcomes were categorised as either short (≤ 3 months), intermediate (> 3 but < 12 months) or long (≥ 12 months) term. The short-term follow-up was considered the primary outcome time point. If multiple time points fell within the same period, the one-time point closest to 7 weeks, 6 months and 12 months for each follow up period was used.
If relevant data were missing, the authors were contacted to request clarification or additional data (e.g. in the case where only an abstract was available (27, 28). If data were not available within the text, means and standard deviations (SDs) were estimated from graphs and figures, if available. If SDs were not reported, we attempted to estimate them from the confidence intervals (CIs) or other measures of variance. If SDs were missing for follow-up outcomes, we used the SD for that outcome at baseline.