This study is a cross-sectional sub-study from a parent prospective cohort study of pulmonary TB patients and their household contacts, conducted between February and November 2017 in Lima, Peru, which was aimed at determining the prevalence of DM and prediabetes (preDM) in persons with TB. The study was carried out in the Public Hospital Sergio Bernales and outpatient health centers of the Carabayllo and Comas districts. People with pulmonary TB ≥18 years of age diagnosed by the Peruvian National TB Program who were not receiving ATT or had started it within a period of no more than 5 days were included. Exclusion criteria were patients or contacts diagnosed with HIV, pregnant people, those who did not live permanently in the jurisdiction area of the study, and patients who had infection or disease due to non-tuberculous mycobacteria.

The follow-up of TB cases and their household contacts was carried out up to 6–12 months after enrollment through three visits (baseline, month 2, and month 6). In each visit, clinical and epidemiological information (e.g., age, sex, and prior TB, among others) was collected. Additionally, specimens such as sputum and blood were also collected for analysis of hemoglobin (Hb), glycated hemoglobin (HbA1c), the oral glucose tolerance test (OGTT), and fasting plasma glucose (FPG) in all patients. The recruitment details and laboratory and field procedures are shown in the Supplementary Figure 1. More information can be found in a previous study with the same cohort (26). In this study, participants with TB-multi drug resistance (MDR) were excluded, and only data collected from the TB cases and their contacts at baseline were analyzed (before the TB cases initiated anti-TB treatment).

The previous prospective study was expected to enroll 130 individuals with TB and 130 contacts, considering 80% of power to a difference in DM prevalence of 10% between the study groups and a significance level of α = 0.05. In this study, 136TB cases and 133 household contacts were included, and a power of 81.9% was obtained (significance level of α = 0.05).

TB cases were diagnosed in the health centers, according to the definitions of the Peruvian National TB Program (27), by one or more of the following criteria: (a) clinical factors (presumptive diagnosis), (b) bacteriology (sputum smear positive) or positive culture (solid or liquid), (c) GeneXpert MTB RIF, and (d) chest radiography. Contacts of TB cases were defined as a person aged 12 years or older, without TB diagnosis, who at least shared the household where they slept or ate their meals.

DM was defined in agreement with the American Diabetes Association (ADA) guidelines (28) as 2 h OGTT≥200 mg/dL, HbA1c ≥6.5%, or FPG ≥126 mg/dL. PreDM was also defined in agreement with the ADA guidelines as 2 h glucose 140 a 199 mg/dL, HbA1c 5.7–6.4%, or FPG 100–125 mg/dL. Dysglycemia was defined as DM or PreDM. In this study, participants were classified into dysglycemia, TB, TB–dysglycemia, and the control group, comprising non-TB/non-dysglycemia people.

Anemia was identified following the WHO criteria as Hb levels below 12 g/dL and 13 g/dL for female and male patients, respectively.

Finally, TB treatment outcomes such as recurrence, treatment modification, treatment failure, and death (during treatment) were classified as unfavorable outcomes. Cure and complete treatment were defined as favorable outcomes. All definitions were established according to the Peruvian National TB Program (27) and are described in the Supplementary Table 1.

The food intake was obtained according to a food frequency questionnaire (FFQ), with 94 food items. The FFQ was specifically developed for this study to investigate and delineate the dietary patterns of people with TB and their household contacts with and without dysglycemia among the Peruvian population and has thus not been validated in previous studies. In addition, the data retrieving and processing of variables were performed following standardized steps from a previous study of our group (24, 25).

The FFQ was applied to TB cases and their household contacts at the enrollment visit (before the patient initiated anti-TB treatment). The consultations on habitual diet referred to the consumption of food outside and inside the home in the last 6 months. The food frequency choices had the following response options: Never / rare; 1 to 3 times a month; 1 time per week; 2 to 4 times a week; and ≥ 4 times a week. In addition, the number of times these foods were consumed was investigated. After data collection, using food composition and nutrition tables, we standardized the amounts consumed of each food and/or preparation, referred to units of weight (g) and/or volume (mL), which were used for calculating the daily consumption of the food recorded in the FFQ. Industrialized foods and / or preparations that were not included in the tables were searched through the Internet directly on the manufacturer’s website or through recipes. Thus, it was possible to obtain an approximation of the total daily food consumption in grams through calculations based on weekly and monthly consumption.

The average daily consumption in grams of these food items was provided by the self-reports of the participants. For the statistical analysis, the food items that composed the FFQ were categorized into 10 food groups according to the similarity in nutritional composition and food habits of the Peruvian population: Sugar and sweets, sweetened beverages, tubers, fast food, oils, milk and dairy, fruits and vegetables, rice and cereals, meat, and legumes. The approach used for creating these groups is described in Supplementary Table 2.

To analyze the potential association between food consumption profiles and anthropometric and dysglycemia status, we used an analytical approach denominated ‘systems nutrology’, created and detailed by our group in the article of Andrade et al., based on the ecological analysis approach (24, 25).

To evaluate the overall consumption profile of the different food groups, we transformed data on the total consumption of each food group (in grams) into the abundance of consumption relative to total diet, or relative abundance (24, 25). After defining the proportion of consumption of each food group in the total ingestion, we performed an unsupervised hierarchical cluster analysis to define the dietary patterns (Ward’s method). The different clusters, named dietary patterns (DPs), were defined by the overall similarity of food consumption, represented using dendrograms denoting the Euclidean distances. We also used bubble plots to illustrate the representativeness of consumption of each food group in each one of the DPs calculated. Pie charts were used to illustrate the frequency of individuals grouped by clinical group, sex, and anemia status in each DP.

Characteristics of the study participants were presented as median and interquartile ranges (IQR) for continuous variables or frequency (%) for categorical variables. The Gaussian distribution of continuous variables was assessed using the Kolmogorov–Smirnov test (no variable was normally distributed) and compared using the Mann–Whitney U test (between two groups) or Kruskal–Wallis test (between more than two groups) with Dunn’s and Tukey’s multiple comparisons. Categorical variables were compared using the Fisher’s exact test (2 × 2 comparisons) or Pearson’s chi-square test (more than two groups). For comparisons of abundance of food consumption, one-way ANOVA was used.

Correlations between total intake in grams of each food group and clinical variables associated with glucose levels (HbA1c and FPG) were evaluated by the Spearman rank test. We created correlation matrices stratifying the participants according to TB status.

To test the association between the unfavorable ATT outcome and DP identified in the hierarchical cluster analysis, we performed a binary logistic regression analysis. The selection of variables was based on Lasso regression. Seventy percent of the cases were randomly selected as the training set, and the rest of the cases were included in the validation set for the internal validation of the model. The DP2 was used as the baseline for the estimation of odds of consumption of the other patterns, given that most patients in this cluster were non-TB/non-dysglycemia. The results were presented in the form of adjusted odds ratio (aOR) and 95% confidence intervals (CIs).

All the analyses and data visualizations were pre-specified. Two-sided p value < 0.05 after adjustment for multiple comparisons (Bonferroni’s method) was considered statistically significant. Statistical analyses were performed using R 3.4.2 (R Foundation) and SPSS 24.0 (IBM statistics). The R packages used were ComplexHeatmap v2.2.0, Hmisc v4.4.1, and ggplot2 v3.3.2.

Our cohort was composed of 269 individuals from Lima, Peru. Among the participants in the study, 85 (31.59%) were considered non-TB/non-dysglycemia, 48 (17.84%) had dysglycemia, 75 (27.88%) had TB, and 61 (22.67%) had TB and dysglycemia. The median age was highest for people with dysglycemia at 50.6 years (IQR: 40.2–61.0), while TB patients had the lowest median age at 25.8 years (IQR: 21.3–31.0, p < 0.001). In this study, there were 137 male and 132 female participants; however, the frequencies of male participants in the investigated groups (dysglycemia, TB and TB–dysglycemia) were higher than that of female participants. In the control group, with non-TB/non-dysglycemia participants, this proportion was inverse (p = 0.002) (Table 1).

The TB–dysglycemia group presented the lowest frequency of illiteracy (Control: 21.2%; Dysg: 16.6%; TB: 62.4%; TBdysg: 9.84%, p = 0.001) and higher frequencies of illicit drug use (Control: 0.0%; Dysg: 2.1%; TB: 10.7%; TBdysg: 15.0%, p < 0.001), alcohol abuse (Control: 28.2%; Dysg: 50%; TB: 49.3%; TBdysg: 55%, p = 0.004), and anemia (Control: 37.6%; Dysg: 20.8%; TB: 66.2%; TBdysg: 82.0%, p < 0.001), as well the lowest level of hemoglobin [Control: 13.2 (12.2–14.2); Dysg: 13.6 (12.8–14.2); TB: 12.6 (11.2–13.4); TBdysg: 11.4 (10.2–12.6), p < 0.001] (Table 1). TB cases presented lower values of BMI [Control: 26.1 (22.9–29.4); Dysg: 29.6 (27.9–33.3); TB: 22.3 (20.4–25.2); TBdysg: 23.0 (21.5–25.7), p < 0.001]; and waist circumference [Control: 88.0 (79.0–94.0); Dysg: 98.0 (93.8–106); TB: 80.0 (74.0–86.0); TBdysg: 84.0 (21.5–25.7), p < 0.001]. The overall comparisons of clinical data according to TB and dysglycemia status are detailed in Table 1.

In the comparisons of food group intake (grams per day), we observed statistical differences for all the food groups between the clinical groups, with p < 0.001 (Table 2). Non-TB/non-dysglycemia individuals had lower intakes of tubers, rice, and cereals and higher intakes of legumes, fruits, and vegetables compared to the other groups. The dysglycemia group had lower intakes of legumes and higher intakes of meat and milk and dairy. The TB participants presented the lowest intakes of milk and dairy, fruits and vegetables, meat, fast food, sugar and sweets, and oils. Finally, the TB–dysglycemia group presented the highest total consumption, with higher intakes of rice and cereals, tubers, fast food, sweetened beverages, sugar and sweets, and oils. The values for each food group according to TB and dysglycemia status are described in Table 1.

An unsupervised hierarchical cluster analysis (Ward’s method) of the abundance of food consumption (which infers the representativeness of the intake of each food group in the overall diet) was used to identify DPs of the consumption of distinct food groups, independent of the study group (non-TB/non-dysglycemia, dysglycemia, TB, and TB–dysglycemia). It was possible to identify three major DPs: DP1 (n = 58), DP2 (n = 70), and DP3 (n = 141) (Figure 1A and Table 2). All TB cases (with and without dysglycemia) were distributed in DP1 and DP3, while no TB patient was grouped in DP2. DP1 was represented mostly by TB–dysglycemia cases (91.4%), while in DP2, there were more frequent control individuals (87.1%). Meanwhile, DP3 was composed mainly of TB cases (51.1%) (Figure 1B and Table 2). While analyzing the DP according to the relative abundance of the food groups consumed, it was possible to observe that the rice and cereals group was the most representative food group in all the three clusters, followed by tubers (Figure 1A). However, in comparing the groups with each other, we observed that DP1 exhibited the highest intakes of these food groups, including sweetened beverages and fast food, and the highest frequency of persons of male sex and with anemia (Figure 1B and Table 2). In contrast, DP2 presented the highest frequency of female participants (p < 0.001) and displayed higher intakes of legumes, fruits, and vegetables, while DP3 presented the highest intakes of milk (Figure 1B and Table 2). Furthermore, DP1 showed a higher median age than DP2 and DP3 (p < 0.001) (Figure 1C and Table 2). Additional analyses testing associations between the DPs and the clinical groups are detailed in Table 2.

In general, DP1 was predominantly composed of individuals who related the use of cannabis (12.3%, p = 0.008) and illicit drugs (14%). In addition, this DP presented the highest values of BMI (23.8 kg/m2, IQR: 21.9–26.5, p < 0.001), FPG (104.2 g/dL, IQR: 99.3–150.4, p < 0.001), and HbA1C (5.6%, IQR: 5.1–6.8, p < 0.001), in contrast with other DPs. In contrast, DP1 presented the lowest Hb levels (11.7 g/dL, IQR: 10.3–13.1, p < 0.001) in comparison with other groups, as shown in Table 2. DP3 was composed mainly of individuals with the highest frequency of illiteracy (28.4%, p = 0.002) and lowest BMI values (23.5 kg/m², IQR: 20.9–27.2, p < 0.001). Finally, DP2 was composed of a small frequency of male participants (40%, p = 0.002), without illicit drug or cannabis use reported, and the highest BMI (27.7, IQR: 26–30.3, p < 0.001), waist circumference (92 cm, IQR: 87–100, p < 0.001), and Hb levels (13.3, IQR: 12.7–14.4, p = 0.014) (Table 2).

To evaluate the influence of the intake of food groups on the glycemic status of the study participants (assessed here through FPG and HbA1c levels), we performed a Spearman correlation test between these two variables and each food group, stratifying the participants according to TB status. In our cohort, we observed that there were higher correlations between some food groups and FPG (Figure 2A), but no statistically significant correlation was observed with HbA1C frequency (Figure 2B). In individuals without TB, consumption of rice and cereals and tubers was positively correlated with FPG levels, while consumption of legumes was negatively correlated with this variable. In TB cases, besides the consumption of rice and cereals, fast food and oils were also positively correlated with FPG levels (Figure 2A).

When we classified the TB population according to treatment outcome, 28 individuals (23%) had an unfavorable outcome [treatment modification (n = 4), loss to follow-up (n = 5), failure (n = 8), and death (n = 11)], while 92 individuals (77%) completed treatment favorably (Supplementary Table 3). TB cases with unfavorable outcomes were characterized by elevated HbA1c levels (5.3% vs. 5.1%, p = 0.004) and FPG (100 g/dL vs. 93.5 g/dL, p = 0.003) compared to TB cases that had a favorable treatment outcome. Consequently, 85.7% of individuals with unfavorable outcomes had dysglycemia, whereas only 30.4% of individuals with a favorable outcome had this condition (p < 0.001) (Supplementary Table 3).

Regarding consumption according to food group, notably, participants who experienced unfavorable outcomes had, at baseline, higher consumption of rice and cereals, tubers, fast food, sweetened beverages, sugar and sweets, and oils. In contrast, individuals with a favorable outcome exhibited a lower intake of the aforementioned food groups (p < 0.05) (Supplementary Table 3).

We performed a second hierarchical cluster analysis among only people with TB using food groups, clinical parameters, dietary patterns (DP1 and DP3, identified in this study subpopulation), and TB treatment outcomes (classified as unfavorable and favorable). This analysis resulted in two major clusters (Figure 3A). Cluster I displayed a profile characterized by significantly greater consumption of all food groups, except legumes (Figure 3A and Supplementary Table 4). Interestingly, in Cluster I, 51% of the study participants experienced an unfavorable outcome, and this percentage was substantially higher than the 4% detected in Cluster II (p < 0.001). Furthermore, Cluster I was mostly composed of people with TB–dysglycemia (86%, p < 0.001) and higher frequency of illicit drug use (19.6%, p = 0.041), medians of age (37.8 years), HbA1c (5%), and FPG (100.8 g/dL) compared to Cluster II. More details about Cluster I and Cluster II are presented in Supplementary Table 4. Notably, participants from Cluster I more frequently had DP1 (73%) than those in Cluster II (12%) (p < 0.001) (Figure 3B). Individuals who had DP1 more frequently experienced unfavorable TB treatment outcomes (49%) when compared with those who had DP3 (8%) (p < 0.001); death was the most frequent outcome in persons who had DP1 (Figure 3C).

Finally, to test for an independent association between the DPs identified and the clinical groups with a risk of developing an unfavorable TB treatment outcome, we performed a logistic regression analysis. Using this approach, we found that DP1 (aOR: 3.53, 95%CI: 1.01–12.34, p = 0.048) and age (aOR: 1.04, 95%CI: 1.01–1.07, p = 0.017) were associated with increased risk of unfavorable TB treatment outcomes, independently of TB–dysglycemia status (adjusted Odds Ratio [aOR]: 4.33, 95%CI: 1.00–18.73, p = 0.050) (Figure 3D).