Statistical analysis

All data cleaning, visualisation, and statistical analysis was performed on R [21]. Numerical variables were described by median and interquartile range (IQR) for age and average household monthly income, and categorical variables were described by proportion (percentage).

Univariable regression was used to first investigate factors associated with patient, diagnostic, and treatment delay and numbers of encounters until diagnosis, and then a multivariate model was fitted to control for confounders and investigate the effect of the COVID-19 pandemic on the outcomes of interest. Age, gender, education level, employment status, average household income, insurance status, comorbidities, minutes to nearest CHC, cough duration, and having a fever were included as risk factors for delays. Additionally, we inquired about all the encounters the individual had with various healthcare providers leading up to their diagnosis. Specifically, the purpose of the visit, the type of provider, and the provider’s sector.

Due to the outliers in the data and the skewed but unimodal underlying distribution of the continuous outcome variables, quantile regression (tau = 0.5) was utilised to examine the association between the outcome of interest (patient delay, doctor delay, treatment delay, number of encounters) and their associated factors. Univariable regression was used to first investigate factors associated with patient delay and numbers of encounters until diagnosis, and then a multivariate model was fitted to control for confounders and investigate the effect of the COVID-19 pandemic on the outcomes of interest. The associated factors for the patient delay model were chosen based on the previous manuscript by Lestari et al [14] and after reviewing literature regarding the most common factors associated with patient delays. Age, gender, education level, employment status, average household income, insurance status, comorbidities were taken from the previous manuscript [19], and minutes to nearest CHC was included as a variable as the association between patient delay and long distances from a healthcare facility has been proven in multiple systematic reviews summarising the risk factors of delays in TB [22–25]. An additional two variables were added that might have confounded the relationship between COVID-19 and TB: the symptoms of cough and fever prompting the initial visit to a healthcare provider. We hypothesized that having a fever or a cough can be associated with delays, as these the presence of these symptoms can prompt an immediate visit to a primary health care provider [26] or can be treated as non-severe symptoms and increase delays/number of encounters to informal providers [23, 24, 27, 28]. Sensitivity analyses to cross-validate findings from the quantile regression were performed by using logistic regression for the same models and are in the supporting information. A cut-off of 30 days was used to indicate patient delay (Table A in S1 Text), and a cut-off of 6 encounters was used to indicate a high number of encounters (Table B in S1 Text), with all the same predictors as mentioned above.