Inflammation Precedes Paradoxical Reactions in HIV-Negative Patients With Pulmonary Tuberculosis

Abstract

Background

“Paradoxical reactions” are a poorly understood phenomenon of clinical worsening during adequate antibiotic tuberculosis treatment. We aimed to characterize and predict paradoxical reactions using clinical, microbiological, and inflammatory factors in human immunodeficiency virus (HIV)–negative patients with pulmonary tuberculosis.

Methods

We closely monitored HIV-negative patients aged ≥16 years who were admitted to a Dutch referral center. Paradoxical reactions were defined as new or increased tachypnea with hypoxemia or new or increased need for supplemental oxygen, after exclusion of alternative causes. Blood inflammatory parameters at onset of paradoxical reaction, and 1 and 2 weeks prior were matched for treatment duration to those of patients without paradoxical reactions and compared using Stouffer z scores. Baseline and longitudinal mycobacterial load, clinical parameters, chest radiographic involvement, and antibiotic plasma concentrations were assessed for association with paradoxical reactions.

Results

Among 81 HIV-negative patients with pulmonary tuberculosis, 19 (24%) experienced a paradoxical reaction, after a median of 13 days of antituberculosis treatment (interquartile range, 8–18 days). Paradoxical reactions were characterized by higher C-reactive protein level (Stouffer z = 4.22), neutrophil count (z = 3.65), neutrophil-to-lymphocyte ratio (z = 4.94), and ferritin level (z = 4.12) and lower lymphocyte count (z = −3.26), monocyte count (z = −4.72), albumin level (z = −6.33), and hemoglobin level (z = −2.32). More extensive lung involvement and higher mycobacterial loads at initiation of antituberculosis treatment were associated with paradoxical reactions. Antituberculosis drug levels were adequate, and mycobacterial loads declined in both groups. High C-reactive protein and neutrophil-to-lymphocyte ratio and lymphopenia at initiation of antituberculosis treatment predicted paradoxical reactions.

Conclusions

Paradoxical reactions in HIV-negative individuals with pulmonary tuberculosis concur with, and are preceded by, outspoken systemic inflammation. Baseline inflammatory parameters may guide future stratified preemptive immunomodulation.

Paradoxical reactions in HIV-negative individuals with pulmonary tuberculosis concur with and are preceded by outspoken systemic inflammation. At baseline, extensive lung involvement and high mycobacterial load and C-reactive protein can identify patients at risk, which may guide stratified preemptive immunomodulation.

Graphical Abstract

Graphical Abstract.

Treatment of tuberculosis can be complicated by a paradoxical reaction, often defined as clinical and/or radiological worsening after initiation of appropriate antituberculosis treatment [1–4]. Paradoxical reactions vary depending on disease localization and can include respiratory deterioration and prolonged fever [4, 5], causing relevant morbidity and use of additional resources.

Paradoxical reactions are named after the counterintuitive parallel occurrence of worsening with adequate antituberculosis treatment and effective mycobacterial killing. They are traditionally managed with corticosteroids [2], and in recent years other immunomodulatory agents have been used, including tumor necrosis factor (TNF) inhibitors [6] and interleukin 1 receptor antagonists [5]. For human immunodeficiency virus (HIV)–associated immune reconstitution inflammatory syndrome (IRIS), high baseline mycobacterial load and poor immunity with swift recovery are known risk factors [7, 8]. Extrapolating this to HIV-negative patients, it is hypothesized that initiation of antituberculosis treatment leads to increased release of Mycobacterium tuberculosis antigens, leading to an exaggerated inflammatory response and increased proinflammatory cytokine production [3–5], but supporting data are lacking. Moreover, underlying mechanisms and risk factors for paradoxical reactions in HIV-negative patients are unclear, especially for those with pulmonary tuberculosis, the most common form of tuberculosis [9].

Therefore, in a cohort of well-characterized HIV-negative patients hospitalized with pulmonary tuberculosis, we examined patient characteristics and inflammatory markers in relation to the development of a paradoxical reaction, as this could aid early recognition and prompt management. First, we profiled laboratory inflammatory parameters of patients during and before paradoxical reactions. As a second objective, we aimed to identify clinical, inflammatory, microbiological, and radiological parameters at the start of antituberculosis treatment that identify patients at risk for a paradoxical reaction.

METHODS

Patient Consent Statement

Patients receiving care for mycobacterial infections at Radboud University Medical Center (Nijmegen, the Netherlands), a national referral center, provided informed consent for inclusion in the Mycobacterial Cohort Study, an ongoing prospective cohort approved by the Medical Ethics Committee Oost-Nederland (CMO NL2021-13231).

Study Design and Participants

HIV-negative patients ≥16 years of age who were hospitalized with pulmonary tuberculosis, with or without concurrent extrapulmonary manifestation, were eligible for the current study. Patients were excluded if they had (1) definite or probable tuberculous meningitis prompting corticosteroid treatment; (2) <3 different microbiologically active antituberculosis drugs for >14 days within the first 2 months after commencing treatment; (3) missing clinical data precluding patient categorization; (4) follow-up totaling <8 weeks; or (5) receipt of TNF inhibitors within <5 half-lives before the start of antituberculosis treatment. Antituberculosis treatment included a protocolized increased dose of rifampicin (>600 mg) in patients with (large) cavities, diabetes mellitus, disseminated disease, or body mass index (BMI) ≤18 (calculated as weight in kilograms divided by height in meters squared) accompanied by significant weight loss.

Outcome Definition

Paradoxical reactions were defined as respiratory worsening after initiation of appropriate antituberculosis treatment, characterized by tachypnea (respiratory rate ≥20/min) combined with hypoxemia (peripheral oxygen saturation ≤95% on pulse oximetry) or need for supplemental oxygen for ≥3 days. The “no paradoxical reaction” group included patients without respiratory worsening, as well as those with respiratory worsening explained by other causes, such as decompensated heart failure or hospital-acquired infections. A researcher (L. K.) identified paradoxical reactions retrospectively, blinded to baseline data, with review by an infectious diseases physician (A. v. L.). Conflicts in designation were solved through discussion. Fever was not included in our definition of paradoxical reactions because inflammatory markers were the main determinants in this study.

Patients already meeting the respiratory criteria at admission, who did not show initial improvement after start antituberculosis treatment, were categorized as having a paradoxical reaction at the moment they showed further worsening: increased tachypnea (respiratory rate ≥4/min higher than at admission) or increased need for supplemental oxygen (≥2 L/min higher or with a switch to high-flow oxygen suppletion or mechanical ventilation). The day of initiation of appropriate antituberculosis treatment, defined as uninterrupted administration of ≥3 microbiologically active antituberculosis drugs, was used as the baseline to calculate the number of days to the day of paradoxical reaction in our retrospective assessment and the day of any clinical diagnosis of paradoxical reaction.

Clinical Data and Blood Inflammatory Parameters

Clinical data were collected from electronic health records. During admission, respiratory rate, peripheral oxygen saturation, oxygen supply, and body temperature were measured 3 times daily in the first weeks of admission and more frequently when clinically indicated.

Blood inflammatory parameters were routinely measured at least once weekly during admission. Ferritin levels were measured retrospectively from weekly biobanked plasma samples. For patients initially admitted to other centers, laboratory values were extracted from referral letters. Anergic responses were defined as either negative or indeterminate responses to interferon γ release assays or—when these assay results were not available—by negative or intermediate responses to the tuberculin skin test. Laboratory measurements were log-transformed if the data were nonnormally distributed, after addition of 0.05 × 10⁹/L for leukocyte differentials to account for 0 measurements, and after replacement with half of the lower limit of quantification for C-reactive protein (CRP) values below that lower limit. For patients who did not experience a paradoxical reaction, measurements were censored at the time they received immunomodulatory drugs.

Imaging

Chest radiographs at the start of antituberculosis treatment, including those received digitally when patients had initially been admitted to other centers, were assessed retrospectively. Two chest radiologists, blinded for patient data including outcome, scored radiographs independently for the presence of consolidative disease, cavitary disease and the size of the largest cavity, miliary disease, lymphadenopathy, and pleural effusion. The percentage of lung involvement was estimated visually as a percentage of total lung tissue. Furthermore, a severity score on a scale of 0–8 was determined by scoring each of 4 quadrants visually as 0 (no involvement), 1 (mild involvement; estimated involvement of 0%–50%), or 2 (moderate-severe involvement; estimated involvement >50%) [10]. Disagreements were resolved through consensus reading.

Therapeutic Drug Monitoring

Therapeutic drug monitoring of antituberculosis drugs (ie, individualization of drug doses based on measurement and interpretation of plasma drug concentrations) was performed ≥14 days after initiation of antituberculosis treatment, when stable drug concentrations (“steady state") for all first-line antituberculosis drugs can be expected. Levels were measured 2, 4, and 6 hours after drug administration according to our clinical practice, and modeled area under the concentration-time curve (AUC) from 0 to 24 hours (AUC_0–24h) values were calculated [11]. Adequate exposure was defined as AUC_0–24h ≥75% of the respective average total exposure in the reference population for all antituberculosis drugs, except for linezolid for which adequate exposure was individually determined based on AUC/bacterial minimum inhibiting concentration ratios >100 [12]. Adequate concentrations of ≥2 individual microbiologically active drugs were required for the treatment regimen to be considered pharmacologically active.

Sputum Time-to-Positivity Modeling

Sputum samples were collected routinely during admission for direct microscopy with auramine staining plus solid and liquid mycobacterial cultures. For patients initially admitted to other centers, sputum samples' liquid culture time-to-positivity were queried. Isolates were identified by means of whole-genome sequencing. Standard procedures included weekly sputum cultures follow-up until clinical improvement, supported by physiotherapy or sodium chloride nebulization if needed.

The log₁₀-transformed liquid culture time to positivity was analyzed using nonlinear mixed-effects models incorporating a linear or bilinear slope to quantify the decrease in mycobacterial load over time. The time to positivity at initiation of antituberculosis treatment and slopes were compared between patients who experienced a paradoxical reaction and those who did not, using a log-likelihood-ratio test (modeling details in Supplementary Methods, Supplementary Tables 1 and 2, and Supplementary Figure 1).

Data Analysis and Visualization

Data were analyzed and visualized using RStudio software (version 2024.04.2) [13]. Cross-sectional comparisons of patient characteristics, radiological assessment, laboratory measures, and plasma concentrations of antituberculosis drugs were performed using Fisher exact tests for nominal variables when contingency table cell counts were <5 and χ² tests otherwise. Continuous variables were compared using Mann-Whitney U tests. Hazard ratios were calculated using univariable Cox regression analyses. For CRP the Youden index (Youden's J statistic) was calculated. Spearman R correlation coefficients were calculated for clinical, radiological, and laboratory parameters at initiation of antituberculosis treatment. Cumulative incidence curves were generated for CRP, lymphocyte count, neutrophil-to-lymphocyte ratio, and ferritin and albumin levels, and differences were assessed using log-rank tests.

For each patient with a paradoxical reaction, blood inflammatory parameter levels were converted to z scores using the distribution of levels from patients who did not experience a paradoxical reaction within the same 7-day window (±3.5 days) relative to the start of antituberculosis treatment. The z scores were combined using the Stouffer weighted z-score method, adjusting for sample sizes.

Locally estimated scatterplot curves were generated for each group to visualize the trajectory of vital and blood inflammatory parameters over time. For vital parameters with >1 measure per day, the geometric mean was used for respiratory rate and oxygen saturation, the mean for oxygen suppletion, and the daily maximum for temperature. The geometric mean of blood inflammatory parameter values was used if >1 value was available within a week.

RESULTS

Of 157 patients with tuberculosis recruited in the Mycobacterial Cohort Study between January 2022 and March 2026, 81 HIV-negative adult patients with pulmonary tuberculosis were included (Figure 1A). Tuberculosis was microbiologically confirmed by means of culture or polymerase chain reaction in all patients, of whom the majority were male (77%), with a median age of 41 years (interquartile range [IQR], 26–55 years). Almost half of all patients (46%) originated from tuberculosis-endemic countries (incidence, >50/100.000/y), and 57% had started antituberculosis treatment in a referring center a median (IQR) of 7 (4–11) days before admission to our center.

Figure 1.

A, Flowchart of patient inclusion, exclusion, and categorization. For the exclusion categories, interrupted treatment was defined as a total of >14 days without ≥3 different antituberculosis drugs in the first 2 months of treatment, missing clinical data prevented patient categorization, and patients with immune suppression were those who received tumor necrosis factor inhibitors within <5 half-lives before the start of antituberculosis treatment. Among the patients with respiratory worsening, 2 had decompensated heart failure, 1 had aspiration pneumonia, and 3 had hospital-acquired bacteremia (coinciding with a tuberculosis-drug adverse reaction in 1, due to pulmonary aspergillus coinfection in 1, and due to pulmonary embolism in 1). B, Clinical course among patients with a paradoxical reaction (PR). Abbreviation: HIV, human immunodeficiency virus.

Following initiation of antituberculosis treatment, 27 patients showed respiratory worsening, and for 19 of 81 (24%) this was due to a paradoxical reaction (Figure 1B). Paradoxical reactions occurred a median (IQR) of 13 ( 8–18) days after initiation of antituberculosis treatment and included (increased) tachypnea in 89% of patients (17 of 19); for the majority (84% [16 of 19]), this led to (increased) oxygen supplementation (Supplementary Table 3). Our retrospective assignment of a paradoxical reactions preceded the clinical diagnosis of paradoxical reaction in 10 patients by a median (IQR) of 5 (3–10) days. In 4 patients, respiratory worsening concurred with the clinical diagnosis; in 2 patients, it followed clinical diagnosis by 1 and 4 days, respectively, while in 3 patients, paradoxical reaction was not clinically considered. Notably, 47% of patients (9 of 19) who experienced a paradoxical reaction had already presented with a respiratory rate ≥20/min and oxygen saturation ≤95% or the need for supplemental oxygen at admission, and only 2 showed initial improvement.

Matched for the duration of antituberculosis treatment, patients with paradoxical reactions had higher blood CRP levels (Stouffer z = 4.22), neutrophil counts (z = 3.65), neutrophil-to-lymphocyte ratios (z = 4.94), and ferritin levels (z = 4.12) and lower lymphocyte counts (z = −3.26), monocyte counts (z = −4.73), hemoglobin levels (z = −2.32), and albumin levels (z = −6.33) than patients without paradoxical reactions. Higher CRP levels, neutrophil-lymphocyte ratios, and ferritin levels and lower lymphocyte counts and albumin levels preceded paradoxical reactions by 1–2 weeks (Figure 2 and Supplementary Table 4). Body temperature was consistently higher in patients with a paradoxical reaction during the first 8 weeks of treatment (Supplementary Figure 2). Paradoxical reactions were managed with corticosteroids, infliximab (TNF inhibitor) or anakinra (interleukin 1 receptor antagonist), and only 1 patient (1%) in the total cohort died during the 8-week follow-up (Supplementary Table 5).

Figure 2.

Heat map of combined z scores for blood inflammatory parameters at the moment of paradoxical reaction (PR) and 1 and 2 weeks prior; z scores show summed differences of the blood parameters of patients with a paradoxical reaction matched to those without, at corresponding times relative to the start of antituberculosis treatment within 1 week (±3.5-day window). The Stouffer method was used to combine multiple z scores. Data availability underlying the analysis is presented in Supplementary Table 4. At 2 weeks before the paradoxical reaction, there were too few ferritin data points to analyze. *P < .05; **P < .01; ***P < .001. Abbreviations: CRP, C-reactive protein; MCV, mean corpuscular volume; MLR, monocyte-lymphocyte ratio; NLR, neutrophil-lymphocyte ratio.

Patients who experienced a paradoxical reaction had lower BMIs than those who did not (Table 1). Comorbid conditions did not differ between patient groups (Supplementary Table 6). Consolidative disease (85% [64 in 75]) and cavitary disease (71% [53 of 75]) were common overall on baseline chest radiographs. Extrapulmonary involvement occurred in 37% of all patients and did also not differ between patient groups. Patients with a paradoxical reaction, however, showed substantially higher chest radiographic severity scores and more lower lung zone involvement.

Table 1.

Patient Characteristics, Microbiological Assessment, Previous Antituberculosis Treatment, and Chest Radiographic Findings

Variable	Patients, No. (%)a		P Valueb
	No paradoxical reaction (n = 62)	Paradoxical reaction (n = 19)
Patient characteristics
Female sex	15 (24.2)	4 (21.1)	>.99
Age, year, median (IQR)	42 (27–56)	40 (25–53)	.68
BMI, kg/m2, median (IQR)c	19.6 (17.3–21.8)	16.6 (15.5–18.1)	<.001
Current or former smoker	32 (51.6)	14 (73.7)	.09
History of previous tuberculosisd	3 (4.8)	3 (15.8)	.14
Comorbid conditions	41 (66.1)	12 (63.2)	.81
Extrapulmonary disease involvement	23 (37.1)	7 (36.8)	.66
Microbiological assessment
Positive sputum auramine staining	41 (66.1)	17 (89.5)	.08
Auramine score (0–4), median (IQR)	2.5 (0–4)	4 (3–4)	.001
Positive sputum culture	42 (70.0)	17 (89.5)	.13
Time-to-positivity, median (IQR), de	8.6 (6.0–13.0)	7.1 (5.5–8.5)	.16
Drug resistance	11 (17.7)	3 (15.8)	>.99
Antituberculosis treatment
Receipt of antituberculosis treatment before admission to center	30 (48.4)	16 (84.2)	.007
Duration of treatment before admission to center, median (IQR), d	7 (4–11)	7 (4–9)	.42
Chest radiographic findingsf
Consolidative disease	47 (82.5)	17 (94.4)	.28
Cavitary disease	39 (68.4)	14 (77.8)	.56
Cavity size, median (IQR), cm	3.8 (3.1–5.2)	5.1 (4.1–6.1)	.11
Miliary disease	1 (1.8)	2 (11.1)	.14
Pleural effusiong	19 (33.3)	2 (11.1)	.08
Lymphadenopathyg	3 (5.3)	2 (11.1)	.59
Severity score (0–8), median (IQR)	3 (1–5)	6 (6–7)	<.001
Proportion of affected lung tissue, median (IQR), %	30 (13–45)	64 (58–75)	<.001
Upper zones involved	46 (80.7)	17 (94.4)	.27
Lower zones involved	36 (63.2)	19 (100)	.002

Abbreviations: ATT, antituberculosis treatment; BMI, body mass index; IQR, interquartile range; PR, paradoxical reaction; TTP, time to positivity.

^aData represent no. (%) of patients unless otherwise specified.

^bFor categorical variables, P values were calculated using the Fisher exact test when contingency table cell counts were <5 and χ² tests otherwise. Differences between continuous variables were assessed using the Mann-Whitney U test.

^cBMI calculated as weight in kilograms divided by height in meters squared.

^dAll patients with recurrent tuberculosis had culture-confirmed tuberculosis.

^eIndividual patient geometric means of time-to-positivity for positive culture results before or within 7 days after antituberculosis treatment initiation.

^fBaseline chest radiographs were available for 57 and 18 patients, respectively.

^gPleural effusion and lymphadenopathy were considered to be present only if independently scored by both chest radiologists, who independently reread anonymized chest radiographs, blinded to patient data, including outcome.

Mycobacterial loads were higher in patients who had a paradoxical reaction. In total, 647 mycobacterial sputum culture results from 80 patients were available for analysis, of which 310 (47.9%) were positive. A bilinear model on the log₁₀ scale described the time to positivity best, with an early fast decline, and a node point of rate switch estimated at 6.2 days (95% confidence interval, 2.8–11 days), followed by a slower decline, in both patients with and those without paradoxical reactions (Figure 3A). The sputum auramine score at initiation of antituberculosis treatment was higher in patients with a paradoxical reaction (Table 1). Consistent with this, the modeled sputum culture time to positivity at initiation of antituberculosis treatment, with more power to detect differences compared with numerical comparisons [14], was shorter in patients who had paradoxical reaction: 6.2 days, versus 9.8 days in those who did not (P = .009). Reassuringly, slopes were similar during the early fast decline and the later slower decline.

Figure 3.

A, Modeled sputum culture time to positivity (TTP) as indication for mycobacterial load in patients with (pink) versus without (blue) a paradoxical reaction (PR). In total, 647 culture results were available, including 310 positive cultures with available time-to-positivity data. Participants with a paradoxical reaction had a higher modeled time-to-positivity at initiation of antituberculosis treatment (ATT) than those without a paradoxical reaction (P = .009). Negative cultures were assigned the maximum time-to-positivity of 42 days. B, Boxplots of total exposure (area under the concentration-time curve from 0 to 24 hours [AUC_0–24h]) to antituberculosis drugs in plasma. Median AUC_0–24h values (interquartile range [IQR]) for patients with or without a paradoxical reaction, respectively, were 166 (102–210) and 102 (50–231) h ⋅ mg/L (P = .15) for rifampicin, 23.5 (8.8–27.7) and 17.0 (9.3–24.5) h ⋅ mg/L (P = .55) for isoniazid, and 284 (276–347) and 372 (292–454) h ⋅ mg/L P = .02) for pyrazinamide. Differences were tested using the Mann-Whitney U test. Population reference values, represented by dashed lines, were 41.1 h ⋅ mg/mL for rifampicin, 15.2 h ⋅ mg/L for isoniazid, and 380 h ⋅ mg/mL for pyrazinamide [11]. Gray areas display exposure levels of antituberculosis drugs <75% the reference value, which were considered inadequate and led to dose increasing. *P < .05; **P < .01; ***P < .001. Abbreviation: NS, not significant.

Therapeutic drug monitoring for first or second-line antituberculosis drugs was performed in 93% of all patients (75 of 81). Regimens were pharmacologically active; that is, they contained ≥2 microbiologically active antituberculosis drugs with total exposure (AUC_0–24h) >75% the population reference value, in 15 of 19 patients (79%) who experienced a paradoxical reaction versus 49 of 56 (88%) who did not (P = .46). Most patients (88% [71 of 81) received a rifampicin-based regimen, and the rifampicin dose was increased (>600 mg) in 68% (48 of 71). More patients with a paradoxical reaction received >600 mg of rifampicin (88% [15 of 17]) vs 61% [30 of 54] in those without a paradoxical reaction [P = .04], equivalent to 34.4 [IQR, 32.9–35.1] vs 18.3 [10.7–34.9] mg/kg body weight [P = .03], respectively. Consequently, the total rifampicin exposure was high, but, importantly, not different between the groups (Figure 3B). The total isoniazid exposure was also similar, while pyrazinamide exposure was lower in patients who experienced a paradoxical reaction. The numbers of patients with rifampicin-free regimens because of resistance (9 of 81 patients) or anticipated toxicity (1 of 81) were too small to analyze separately.

Paradoxical reactions could be predicted by clinical characteristics and laboratory values at initiation of antituberculosis treatment. Clinical predictive characteristics included lower BMI, higher chest radiographic severity score, and higher auramine score (Table 2). Baseline body temperature as a proxy for systemic inflammation was available by semicontinuous monitoring only for patients who had started antituberculosis treatment in our center: 57% of patients (4 of 7) with a subsequent paradoxical reaction had a body temperature ≥38.0°C versus 24% of those (9 of 37) without such a reaction (P = .17). Patients with a paradoxical reaction, had higher CRP levels, neutrophil-to-lymphocyte ratios, and ferritin levels and lower lymphocyte and monocyte count and albumin levels at the start of antituberculosis treatment (Supplementary Figure 3). Interferon γ release assays or tuberculin skin tests showed anergy in 50% of patients (6 of 12) who experienced a paradoxical reaction and 19% of those who did not (7 of 36; P = .04) (Supplementary Figure 4).

Table 2.

Associations of Patient Characteristics, Microbiological Assessment, Chest Radiographic Severity, and Blood Inflammatory Parameters at Initiation of Antituberculosis Treatment With Later Paradoxical Reactions

Variable	Median Value	Proportion With Paradoxical Reaction, No./Total (%)		HR (95% CI)	P Value
		Median or Below	Above median
Age	41 y	11/41 (27)	8/40 (20)	0.71 (.29–1.78)	.47
BMIa	18.8 kg/m2	15/41 (37)	4/40 (10)	0.23 (.08–.70)	.01
Auramine score (0–4)	3	5/41(12)	11/29 (38)	3.72 (1.29–10.72)	.02
Sputum culture time-to-positivityb	8 d	10/28 (36)	5/28 (18)	0.41 (.14–1.20)	.10
Chest radiographic severity score (0–8)	4	1/40 (3)	17/35 (49)	27.2 (3.61–204)	.001
Hemoglobin	7.0 mg/L	9/35 (26)	7/33 (21)	0.78 (.29–2.10)	.63
MCV	84 fL	7/35 (20)	7/31 (23)	1.12 (.39–3.19)	.84
Leukocyte count	8.5 × 109/L	8/34 (24)	8/34 (24)	1.03 (.38–2.74)	.95
Neutrophil count	6.1 × 109/L	6/28 (21)	8/28 (29)	1.42 (.49–4.09)	.52
Monocyte count	0.67 × 109/L	8/29 (28)	5/28 (18)	0.60 (.20–1.84)	.37
Lymphocyte count	1.11 × 109/L	11/29 (38)	3/29 (10)	0.25 (.07–.88)	.03
NLR	6.1	1/28 (4)	13/28 (47)	15.7 (2.05–120)	.008
MLR	0.67	4/29 (14)	9/28 (32)	2.42 (.75–7.87)	.14
CRP	87 mg/L	3/35 (9)	13/34 (38)	5.44 (1.55–19.1)	.008
Ferritin	604 µg/L	0/16 (0)	5/16 (31)	c	c
Albumin	27.5 g/L	14/30 (47)	0/30 (0)	c	c

Abbreviations: BMI, body mass index; CI, confidence interval; CRP, C-reactive protein; HR, univariable Cox regression hazard ratio; MLR, monocyte-lymphocyte ratio; NLR, neutrophil-lymphocyte ratio; PR, paradoxical reaction.

^aBMI calculated as weight in kilograms divided by height in meters squared.

^bA higher sputum culture time-to-positivity was used as a proxy for a lower mycobacterial load.

^cHR and P value could not be calculated due to quasi-complete separation; all patients with a paradoxical reaction had ferritin levels above and albumin levels below their respective medians.

In univariable cox regression analyses, higher CRP levels and neutrophil-to-lymphocyte ratios and lower lymphocyte counts predicted the development of a paradoxical reaction. Log-rank tests (used instead of Cox regression because of quasi-complete separation) indicated a higher risk for paradoxical reactions with higher ferritin and lower albumin levels (Supplementary Figure 5). Multicollinearity between the clinical characteristics and blood inflammatory parameters predicting the development of a paradoxical reaction (Figure 4A), together with the total number of patients with such a reaction, precluded multivariable analysis. CRP may be useful as a single biomarker with easy application, and its values were consistently higher over time (Supplementary Figure 6). The optimal discriminative CRP cutoff between patients who will and those who will not experience a paradoxical reaction was 85 mg/L, with an associated hazard ratio of 8.23 (95% confidence interval, 1.87–36.25; P = .005) (Figure 4B).

Figure 4.

A, Correlation matrix showing Spearman R correlation coefficients between clinical, radiological, and laboratory parameters at initiation of antituberculosis treatment (ATT). *P < .05; **P < .01; ***P < .001. B, Cumulative incidence of the development of a paradoxical reaction, stratified by C-reactive protein (CRP). Differences were assessed using log-rank tests. Abbreviations: BMI, body mass index; MCV, mean corpuscular volume; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio; TTP, time to positivity.

DISCUSSION

In this prospective cohort of HIV-negative patients with pulmonary tuberculosis in the Netherlands, we show that outspoken inflammation accompanies, and precedes, paradoxical respiratory worsening following initiation of antituberculosis treatment. Patients with a paradoxical reaction have higher CRP levels, neutrophil-to-lymphocyte ratios, and ferritin levels and lower lymphocyte counts, monocyte counts, and albumin levels, as well as lower BMIs, more extensive lung pathology, and higher bacterial loads at initiation of antituberculosis treatment.

Dense vital parameter data showed that most patients with paradoxical reactions experienced them within 2 weeks after initiation of antituberculosis treatment, earlier than previously reported (median of 26 days) [15]. This may be explained by our focus on pulmonary tuberculosis, which is associated with earlier paradoxical reactions than extrapulmonary tuberculosis [4], and because we used a sensitive retrospective assessment of respiratory parameters to identify paradoxical reactions. Interestingly, the occurrence or size of cavities was not associated with the occurrence of paradoxical reaction. The extent of lung involvement at initiation of antituberculosis treatment, however, was much higher in patients with a paradoxical reaction, and a subgroup of our patients already met the respiratory criteria at that time point. Importantly, this was followed by improvement in only a few patients but subsequent worsening in all. These patients would not have met the criterion of “initial improvement” required in narrower definitions of paradoxical reactions used in some reports [4, 16]. We argue that using initial improvement as a criterion compromises identification of early paradoxical reactions, as observed in some patients in our setting.

To avoid confirmation bias, fever was not one of our criteria for paradoxical reactions, as we hypothesized that inflammation was important. Body temperature was indeed higher in patients with paradoxical reactions throughout admission, and inflammatory markers were markedly elevated. We propose an important role for the innate immune system in paradoxical reactions, as an increased CRP level can be seen in interleukin 1β/interleukin 6–driven inflammation and an increased ferritin level in interleukin 1β/interleukin 18–driven inflammation [17], concurring with increased neutrophil-to-lymphocyte and monocyte-to-lymphocyte ratios, but this needs confirmation with transcriptional and proteomic profiling. These increased ratios are partially explained by lymphopenia, a baseline risk factor for paradoxical reactions [18], which we also observed in our cohort. We moreover confirm a decreased baseline albumin [15] in patients with a later paradoxical reaction, which likely indicates a prolonged inflammatory state, co-occurring with wasting (low BMI), which has been hypothesized to be TNF driven [19]. Our findings may explain the association between high baseline inflammation and subsequent death [20, 21].

Patients with paradoxical reactions had higher sputum mycobacterial loads at initiation of antituberculosis treatment. The modeled fast load reduction in the first week after the start of antituberculosis treatment substantiates the hypothesis that increased antigen exposure contributes to the pathophysiology of paradoxical reaction. This hypothesis of “hypersensitivity” to Mycobacterium tuberculosis antigens with effective tuberculosis treatment [22] derives from findings in HIV-associated tuberculosis, in which the occurrence of IRIS is associated with smear positivity (as a proxy for high mycobacterial load) [7] and with starting antiretroviral therapy early during antimycobacterial therapy, when mycobacterial loads are presumably highest [2]. Moreover, killed mycobacteria can no longer actively use their immune evasion strategies [23]. We observed higher rates of anergy in patients with paradoxical reactions. Concurring with their notable inflammation, this is in line with an “endotype” of tuberculosis that shows simultaneous hyporesponsiveness and hyperinflammation [24] and is also observed in tuberculosis-associated hemophagocytic lymphohistiocytosis [25]. Future research is needed to understand possible causality between inadequate lymphoid immunity, increased tuberculosis antigens, and a hyperinflammatory myeloid response.

The occurrence of paradoxical reactions, could not be explained by inadequate drug exposure, as <25% of patients in both groups did not meet the standard of pharmacologically active drug regimens: ≥2 antituberculosis drugs with total plasma exposure of ≥75% the population reference value. Focusing on individual drugs, paradoxical reactions were associated with lower exposure to pyrazinamide. Clearance of pyrazinamide increases during the first month of antituberculosis treatment [26], most likely because of rifampicin interaction. It is not known whether more inflammation, as seen in paradoxical reactions, further increases clearance. It is hard to identify a causal role for lower pyrazinamide in the occurrence of paradoxical reactions, as pyrazinamide has only weak early bactericidal activity [27] and pyrazinamide monotherapy is linked to increased lesional ¹⁸F-fluorodeoxyglucose positron emission tomographic activity [28]. Conversely, isoniazid, with strong early bactericidal activity [27], could be hypothesized to contribute to the occurrence of paradoxical reactions, but exposure was not different.

Rifampicin, with modest early bactericidal activity [14], was administrated at an increased dose (>600 mg daily) in most patients. Consequently, rifampicin exposure overall was higher compared with reference populations [11], but again, not different between patients with and those without a paradoxical reaction. Concluding, in contrast to findings in 4 patients with tuberculous meningitis [29], we did not find inadequate drug exposure associated with paradoxical reactions, further supported by similar declines in bacterial load in patients with and those without a paradoxical reaction.

Our study is the first to use densely observed vital parameters to identify paradoxical reactions and to combine longitudinal laboratory data to characterize paradoxical reactions in HIV-negative patients with pulmonary tuberculosis. Culturing of consecutive sputum samples and therapeutic drug monitoring excluded antibiotic treatment failure as a cause of excessive inflammation. Our retrospective analysis of prospectively collected vital parameters to determine the onset of respiratory worsening allowed for early recognition of paradoxical reactions. Limitations include the limited sample size, the observational nature of this study, the risk of case ascertainment, and the selection of more severely ill patients seen in our setting as a national referral center for mycobacterial infections, which precludes statements on paradoxical reactions incidence and generalizability to tuberculosis populations managed by public health services. Replication in cohorts in other settings is warranted. Because vital parameters were sparsely available before admission to our center, earlier deterioration may of course have been missed. Limited data availability for ferritin warrants more careful interpretation of these results. Notably, patients with HIV coinfection were not included in this analysis because HIV-associated IRIS has antiretroviral therapy–induced CD4⁺ T-cell recovery as a known driver of inflammation.

In conclusion, paradoxical reactions in HIV-negative patients with severe pulmonary tuberculosis and high mycobacterial load are characterized by outspoken inflammation. Even at initiation of antituberculosis treatment, high CRP levels, neutrophil-to-lymphocyte ratios, and ferritin levels, and low lymphocyte counts and albumin levels may help identify those at risk of a paradoxical reaction, possibly allowing for early recognition and stratified initiation of immunomodulatory treatment.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.