Analysis of loss to follow-up in 4099 multidrug-resistant pulmonary tuberculosis patients

Discussion

Incomplete treatment of MDR-TB is an important driver of continued transmission and avoidable TB morbidity and mortality worldwide. Using the largest dataset of individual MDR-TB patients currently available, our study is the most comprehensive assessment to date of the timing and risk factors for LFU. The overall frequency of 17.1% is slightly lower than reported in the full dataset (23%) [1] and much lower when deaths are excluded from global TB monitoring at the time (28.6%) [5, 19]. This substantial difference is likely to be due to the countries included in our study. Of note, large numbers of MDR-TB patients in India, China, Kazakhstan and one cohort from South Africa are not included, and this may limit the generalisability of our findings The median (IQR) time to LFU was 7 (3–11) months. Subgroups with a higher risk of LFU were those aged 36–50 years (versus 0–25 years), those HIV positive (versus HIV negative), those receiving a standardised treatment regimen (versus an individualised regimen) and those with no reported serious adverse event (versus reported serious adverse event). The majority of LFU occurred in the initial stage of treatment, with 75% of all LFU occurring in the first 11 months of treatment.

Our results demonstrate that individualised treatment regimens may prevent patients from becoming LFU, although it should be noted that the definition of standardised regimens has changed over recent decades and so this may not be a consistent comparison across cohorts. An individualised approach is challenging to scale-up in many settings as it depends on robust drug sensitivity testing and the clinical skills to alter regimens accordingly. From a programmatic perspective, the policy decision about which model of treatment to pursue should consider LFU as a key factor. This may also have relevance for the scale-up of standardised 9–11-month regimens that the World Health Organization now promotes [20].

The timing of LFU in our study (comprising treatment prior to 2008) provides findings from the recent past that are broadly similar to studies reporting more recent treatment cohorts (published during 2006–2015) [7–16], where the majority of LFU tended also to be in the initial intensive phase of treatment (see Introduction). Further research, similar to our study, on aggregated datasets from more recent cohorts (and particularly cohorts using the new 9–11-month regimens) would help to identify if this pattern of the majority of LFU occurring in the early stages of treatment has continued. As research in this area grows, our findings form an important historical context.

Our analysis shows that having a co-infection of HIV is a risk factor for LFU. Historically, poor access to antiretroviral treatment and the lack of coordination between HIV and TB treatment programmes have been highlighted as factors that lead to worse outcomes for those co-infected [21]. Treatment approaches in HIV and TB differ significantly, which can lead to confusion for patients seeking to navigate these divergent systems of care [22]. In some cases this leads to preferential adherence of HIV treatment over MDR-TB treatment due to increased tolerability and commitment to HIV treatment and a reduced pill burden [23]. This can create further barriers to effective coordination of treatment with patients comorbidly infected with HIV and MDR-TB.

Better treatment outcomes in children compared with adults have been reported in co-infected HIV populations [24] and non-HIV populations [25]. Our analysis identified an increased risk of LFU in those aged 36–50 years (versus 0–25 years). This could be explained by increased family support offered to younger family members with MDR-TB. Furthermore, the competing demands of employment and dependents for individuals in the working-age groups could also be influential in their greater risk of LFU, although an increase in LFU was not seen on those aged 26–35 years [24].

The finding that patients with a reported serious adverse event had an decreased risk of LFU appears counter-intuitive, as one might predict that these experiences increase the likelihood that patients will stop tolerating treatment and disengage. Indeed, previous studies identified that serious adverse events increased the risk of LFU [10, 13]. However, a plausible counter-argument exists that those with a recorded serious adverse event are those who are taking medication, and therefore are more likely to experience a serious adverse event due to their greater adherence and have it recorded due to their engagement with services. This strong engagement with services may reduce their likelihood of becoming LFU. The differing findings of previous studies and our findings here could also be explained by differing recording practices of adverse events.

For several variables in our Cox proportional hazards model (HIV status, additional resistance to pyrazinamide, additional resistance to streptomycin and presence of serious adverse event), the “unknown” category was significantly associated with being LFU. Weaker TB treatment programmes may be the confounding factor, where poorer recording practices and limited efforts at following up patients are both associated with weaker programmes.

The study has several limitations. The large number (55%) of excluded patient records may have introduced bias in our findings. Excluding data with an outcome of death could lead to an overestimation as those that died are not counted in the denominator. Missing data could lead to a potential underestimation of LFU where these missing records may be due to LFU but not recorded as such. Excluding data of those not receiving even the first month of treatment leads to an underestimation of LFU as they are not counted in those identified as LFU. This could also have led to an artefactual prolongation of LFU as patients who would have been early interrupters were selectively removed from the cohorts. The direction of bias from the excluded cohorts is unclear, although the overall rate of LFU is lower than in the main dataset, indicating our overall rate may be an underestimate. The data quality was variable, although attempts have been made to ensure the dataset is as complete as possible. It is possible there are artefactual influences on our dataset that we cannot identify. In addition, we did not have data available for some patient variables that we were interested in, such as comorbid substance misuse or treatment interruptions of <2 consecutive months [26]. The cohorts also do not contain valid measures of the practice of directly observed treatment, which is closely associated with treatment outcomes among MDR-TB patients [27]. Furthermore, there are likely to be other types of programme-related or treatment-related risk factors associated with LFU that we did not have data for, such as the efforts of programmes to engage with patients with interrupted treatments.

Given the high proportion of excluded data, caution is needed when generalising from our findings. In addition, the cohorts we included all pre-date 2008 so these findings may not generalise to current programme management of MDR-TB, particularly for patients treated with the recent standardised 9–11-month regimen [20]. This shorter regimen has the potential to reduce LFU, although since 75% of LFU in our dataset occurred by 11 months, LFU may only reduce by a quarter if the timing of LFU is similar to that in our dataset.

Although some treatment cohorts in this dataset were from resource-constrained contexts, none were from low-income countries. Our findings may not therefore generalise to these settings. We have identified risk factors that are associated with LFU; further work needs to be done to explore the mechanisms that drive stubbornly high rates of LFU in all MDR-TB programmes, including in low-income settings. For instance, the presence of comorbid depression may reduce adherence in MDR-TB as it does in other health conditions [28, 29].

One potential use of the findings of our study for treatment programmes is the development of an evidence-based risk assessment tool for identifying patients at increased risk of LFU. Findings from our study and other aggregated or local studies could inform the development of such a tool. The consideration of certain patient-related and programmatic risk factors (such as those identified in this study) could assist in the identification of patients at high risk of becoming LFU during treatment and prompt proactive interventions to prevent this from occurring.