The effect of early versus late treatment initiation after diagnosis on the outcomes of patients treated for multidrug-resistant tuberculosis: a systematic review

The full original protocol and PRISMA checklist [4] (Additional file 1: Table S1) are available in the supporting information additional file section. The following amendments or clarifications were made to the original protocol: searching Google Scholar in place of Google, including a limit of publication since 1990, including a 10 % check of sifting by a third reviewer, and allowing a flexible cut-off during sifting in the definition of the timing of early versus late treatment.

The research question was framed using PICOT (Population, Intervention, Comparison, Outcome and Time) methodology [5, 6]. The population under consideration was all patients with multidrug-resistant or extensively drug-resistant tuberculosis bacteriologically confirmed by phenotypic methods, or for whom molecular testing indicating rifampicin resistance was used as a surrogate for initiation of MDR treatment. The intervention and comparator of interest were early versus late treatment, defined respectively as drug treatment initiated within 28 days versus later than 28 days after diagnosis of MDR-TB. The primary outcomes in this review were the WHO-defined tuberculosis outcomes as proposed by Laserson et al. [7]: cure, treatment completion, failure/relapse, transfer out, abandoned treatment and death. Treatment success was defined as patients meeting the outcome definition of cure or treatment completed. Poor outcomes were defined as patients with the outcome of failure/relapse, transfer out, abandoned treatment or death. Secondary outcomes pre-defined in the protocol were not included in the search terms, but were intended for consideration during data extraction from included papers, and were defined as adverse reactions from TB drugs (severity, type, organ class), adherence to treatment, or treatment interruption due to non-adherence.

A comprehensive search strategy was developed in consultation with WHO technical experts using the PICOT question as a framework (see Additional file 1). Due to the relatively low number of total hits in preliminary searches, only population and intervention terms were used in the search strategy (Additional file 1: Table S2). By checking for comparator and outcomes defined a priori during the manual sift instead of in the search strategy, the likelihood of missing a potentially relevant paper was reduced. An example of the search strategy, in this case as applied in PubMed, is included in Additional file 1: Table S2.

Electronic health care databases, sources of evidence-based reviews, guidelines, and grey literature were searched in accordance with the specifications of each database. These included PubMed (including MEDLINE), EMBASE, Cochrane library (includes CENTRAL, CDSR, DARE and HTA databases), WHO Global Index Medicus (includes LILACS, WPRIM, IMSEAR, IMEMR, AIM, SciELO and WHOLIS indexes), WHO portal of clinical trials, OpenSIGLE, International Union of Tuberculosis and Lung Disease conference abstracts (2004–2014) and Google Scholar (limited to the first ten pages). The search strategies were executed on 26th September 2015. The date, human studies and language inclusion criteria were applied in the manual sifting process instead of through the limits function of PubMed (and other databases) in order to avoid exclusion of papers that had not been indexed on these criteria (Table 1).

Following removal of duplicates, two-stage screening against inclusion and exclusion criteria was executed independently by two reviewers (JN, AM), sifting first by title and abstract, followed by full-text sifting. A third reviewer (DM) reviewed any discrepant results and every tenth reference to adjudicate and check for consistency. Studies including participants of any age with confirmed multi-drug resistant or extensively drug-resistant pulmonary tuberculosis were potentially eligible for inclusion. Any consecutive case series, case control study, cohort study, randomised controlled trial, systematic review or meta-analysis that included a comparator group was considered for inclusion. Included studies were required to report data on at least one of the primary outcome measures of interest. Although the original intervention of interest was defined as presumed adequate treatment initiated ≤28 days after diagnosis, and a comparator group with treatment initiated >28 days after diagnosis, the protocol was changed to allow flexibility in the definition of this cut off for early versus late treatment. Therefore studies reporting outcomes for patient groups with different definitions for delay to treatment initiation were considered for inclusion.

Any systematic review superseded by an updated systematic review, or narrative reviews not adding new data or new analysis to the existing evidence base were excluded. Finally, studies not performed in humans, written in a language other than those listed in Table 1, or with fewer than ten participants were excluded.

Sifting was primarily managed within Endnote® X7.4 (Thomson Reuters, California, USA). If a paper was deemed ineligible for inclusion at the full text sifting stage, the primary reason for ineligibility was recorded by the reviewers.

An additional post-hoc review of those articles included in the full text review was undertaken by one reviewer (DM) using alternative, less-stringent inclusion criteria. Specifically, papers were identified which reported on two or more groups of patients with different times to MDR treatment initiation. Data on treatment outcome according to WHO criteria or on intermediate outcomes such as culture conversion time (though not validated as surrogates of treatment outcome) were abstracted from papers included in this post-hoc review.

Although the full text review failed to deliver any publications fulfilling the inclusion criteria it was noted that a number of articles reported less well defined data on treatment delay (without a 28 day cut-off) related to some interim outcome measures (smear and culture conversion) and final treatment outcomes [11, 13, 22, 31, 37, 38, 45, 46, 48, 50, 51, 61, 63, 64, 66, 68].

None of these articles addressed the independent effect of treatment delay with a meaningful comparator group - whether not delayed or less delayed - upon treatment outcomes, whether interim or final.

Whilst outside the designated protocol we proceeded to abstract these data in case the resulting narrative yielded any information of use. These data and narrative summary are reported in Table 2 and Additional file 1: supplementary text.