Abstract
Rationale
Undertreatment of tuberculosis (TB) is common in resource-limited settings where same-day microbiological diagnosis is unavailable.
Objectives
We evaluated if a simple clinical risk score for predicting active TB could facilitate treatment initiation among individuals at high risk for TB.
Methods
We conducted a pragmatic, implementation-effectiveness study in periurban primary health clinics in Uganda. Four intervention clinics were paired with standard-of-care comparison clinics. Providers in intervention clinics were trained to use the score and set a threshold score for considering same-day treatment initiation; treatment decisions were at the providers’ discretion. Our primary effectiveness outcome was the change in the proportion of individuals with confirmed TB who started on treatment within 7 days of presentation, comparing preintervention and postintervention periods.
Results
Among intervention clinics, 594 (83%) of 720 people diagnosed with TB started treatment within 7 days during the preintervention period versus 264 (84%) of 316 after implementation (pre–post difference, 1%; 95% confidence interval [95% CI], −6%, 8%). In comparison clinics, 7-day treatment initiation changed from 312 (86%) of 363 preintervention to 153 (73%) of 211 postintervention (pre–post difference, −13%; 95% CI, −22, −5%). A difference-in-differences estimate was 14% (95% CI, 10%, 19%). In intervention clinics, 1,206 (66%) of 1,826 people presenting with TB symptoms were administered the risk score. Two hundred twenty-nine (19%) had a score above the treatment threshold, and 105 (46%) initiated treatment on the same day.
Conclusions
An easy-to-use clinical risk score did not increase 7-day empiric treatment initiation in intervention clinics. However, it improved rapid treatment initiation relative to clinics using the prevailing standard of care. The score was also highly acceptable to clinical providers.
Clinical trial registered with www.clinicaltrials.gov (NCT 05122624).
Keywords: prediction, clinical risk score, tuberculosis, implementation
Tuberculosis (TB) remains the leading single-agent cause of infectious mortality, causing 1.3 million deaths worldwide annually (1). In low- and middle-income countries, people with TB wait an average of 14 days from presentation to TB diagnosis and another 14 days from diagnosis to treatment initiation (2). Delays in diagnosis and treatment are exacerbated by a lack of access to same-day diagnostics or specialists with expertise in TB diagnosis (3–5). Although rapid molecular tools such as Xpert MTB/RIF (Cepheid, Inc.) are universally available, low- and middle-income settings still have logistical challenges (e.g., difficulties in machine maintenance, restricted operating hours, and lack of trained laboratory technicians) (6–9), and many rely on a “hub-and-spoke” model (10), which often makes specimen transport or result return unreliable (9). Indeed, although treatment for TB is deferred until molecular testing is performed and the corresponding result is returned, loss to follow-up before initiation of treatment is common—as high as 20% in Uganda (11, 12). Individuals who are lost to follow-up are at risk for poor outcomes and may transmit TB further within the population (13, 14).
In this context, health facilities without access to same-day molecular testing may benefit from clinical risk scores, which help in decision making about whether to start treatment empirically so that patients with TB are not lost to follow-up while awaiting test results (12, 15, 16). We previously validated a simple risk score for this purpose (17). This score (“PredicTB score”) consists of six individual characteristics (sex [1 point for male], age [1 point for 25–44 yr], HIV status [2 points], diabetes status [1 point], number of symptoms [up to 4 points], and symptom duration [1 point if >14 d]) and is easily calculable as a numeric score between 0 and 10 using pen and paper. The score estimates a patient’s risk of TB on the basis of local levels of Xpert positivity. Same-day treatment can then be initiated for high-risk patients while waiting for microbiological confirmation. We sought to evaluate the implementation and effectiveness of this score as integrated into routine clinical practice in Uganda.
Methods
Study Setting
We identified 12 primary health clinics across six periurban districts within 150 km of Kampala that met four additional criteria: maintenance of high-quality TB records (based on direct review of clinic registers), at least 100 confirmed cases of TB in the preceding year, no access to on-site Xpert testing at the time of study initiation, and willingness of the clinic in charge to have staff trained in using the PredicTB score. We selected eight clinics that were located far enough apart from each other to minimize contamination. We then created four pairs of clinics on the basis of similar geography and patient volume. Within each pair, we randomly assigned one to the intervention arm and one to the standard-of-care arm.
During the course of the study, two intervention clinics and one standard-of-care clinic initiated on-site Xpert testing, rendering them ineligible for the study. These clinics were replaced (at the time of initiating on-site Xpert testing) with another eligible clinic that was most closely matched on geography and patient volume. Thus, although 11 clinics contributed to analyses in total, only 8 clinics were actively participating at any given time (see Table E1 in the data supplement).
Study Procedures
This study consisted of a preimplementation phase (784 d, on average; Table E1), an implementation period (281 d, on average), and postimplementation surveys. In the preimplementation phase, we abstracted data from TB presumptive and treatment registers, including date of clinic presentation, date of sputum submission, type of TB testing, test results, and date of treatment initiation. Each of the study clinics offered Xpert testing as a send-out to a central facility using a “hub-and-spoke” model and also had access to sputum smear microscopy, urine lipoarabinomannan (LAM) testing, and chest radiography. We therefore also abstracted data on sputum smear microscopy, urine LAM, and chest radiography results when available. Microbiologically confirmed TB was defined as a positive result on Xpert testing, sputum smear microscopy, and/or urine LAM.
At the start of the implementation phase, we met with clinical staff, including the in-charge physician, in each clinic to describe the PredicTB score. We then reviewed records at the clinic to estimate the proportion of people being evaluated for TB who had a positive test result and thus the predicted risk of TB for patients with each possible score. We then used this information to motivate a discussion regarding the appropriate score threshold for considering empiric treatment; the final decision regarding this threshold was left to the in-charge physician at each clinic. Once the clinic-level threshold was set, we provided score sheets and decision tables for use in each clinic (Figure E1) and provided a 1-day training to all providers who interact with patients who might have TB. We emphasized that all treatment decisions should be at the provider’s discretion. After this initial training, as a pragmatic study (18), we limited interaction with the study clinics to quarterly visits in which we addressed any concerns, abstracted additional data, and directly observed how the PredicTB score was used among the providers. During these visits, we provided performance feedback to providers; this feedback centered on the proportion of patients for whom the score was used, not on treatment decisions as a result of the score. The filled-in PredicTB score sheets were also collected for further evaluation. In standard-of-care comparison clinics, we performed initial training on Ugandan standards for TB management and likewise conducted quarterly visits to address concerns regarding TB management and the abstraction of TB-related data. During these visits, our field team asked clinic staff about any facility-level changes, trainings, or events that might affect study outcomes. The one event that was a nationwide screening campaign (CAST-TB [Community Awareness, Screening, Testing, and Treatment for Tuberculosis] [19]) was periodically performed in the catchment areas of the study facilities. Individuals identified with TB through participation in CAST-TB were registered with different codes, which enabled us to exclude these individuals from our analyses, because the PredicTB score was validated only for people presenting to clinics for symptomatic evaluation (17). Other than these campaigns and the implementation of on-site Xpert testing (which resulted in clinic replacement, as above), no other major changes or events were noted by clinic staff during these site visits. We also performed TB culture on split sputum specimens from individuals who were empirically diagnosed on the basis of PredicTB score but for whom Xpert results were negative or missing. Last, following the completion of the implementation phase, we surveyed all providers with Likert-type questionnaires who were engaged in TB diagnosis and management in the four ongoing intervention clinics.
Outcomes
We evaluated effectiveness and implementation outcomes following the RE-AIM (reach, effectiveness, adoption, implementation, and maintenance) framework (20). Our primary effectiveness outcome was the absolute difference in the proportion of individuals with microbiologically confirmed TB who were treated within 7 days of initial presentation, comparing the preimplementation period with the postimplementation period in those clinics assigned to the intervention. A key secondary effectiveness outcome was the difference in differences, comparing the primary outcome with the corresponding difference among the standard-of-care comparison clinics. Beyond treatment initiation, we further compared treatment outcomes in the pre- and postimplementation periods in both intervention and comparison clinics. Implementation outcomes included reach (the proportion of people evaluated for TB for whom the PredicTB score was administered), adoption (provider uptake of the PredicTB score), implementation (same-day treatment initiation, 7-d treatment initiation among all people starting treatment for TB [i.e., not only those with confirmed TB]), and maintenance (the proportion of individuals treated empirically, rather than on the basis of laboratory results, over time).
Statistical Methods
Categorical variables were summarized as frequencies; continuous variables were reported as means with 95% confidence intervals (95% CIs). The difference in differences of the time to treatment initiation was evaluated using mixed-effects linear regression with random slopes for each clinic, with a product term between time (pre- vs. postimplementation period) and intervention (intervention vs. comparison arm). The difference in differences of the proportion of individuals with TB who started treatment (same-day or within 7 d) was similarly evaluated using mixed-effects logistic regression. We also estimated the positive predictive value, sensitivity, and specificity of the PredicTB score relative to microbiological confirmation at different PredicTB cutoffs. All analyses were conducted using R version 4.3.2.
Ethical Considerations
The study was approved by the Johns Hopkins School of Public Health Institutional Review Board and the Research Ethics Committee of the Makerere University School of Public Health. Both institutional review boards offered a waiver of informed consent for review of clinical records as representing minimal risk. Written informed consent was obtained from all providers who participated in the direct observation, feedback process, and postimplementation surveys. The study was registered with clinicaltrials.gov (NCT 05122624).
Results
Individuals presenting for TB diagnosis did not differ in terms of age, sex, or HIV status between the preimplementation and implementation periods (Table 1). After the implementation of the PredicTB score, a larger percentage of these individuals submitted sputum for TB evaluation at intervention clinics (74% vs. 56%); however, the proportion of individuals diagnosed with microbiologically confirmed TB remained similar (8–9%) across intervention and standard-of-care comparison clinics both pre- and postintervention. Of 1,826 individuals who were evaluated for TB in intervention clinics during the study period, 1,206 (66%) had the score administered (reach), of whom 229 (19% of those scored) had a score exceeding the clinic-level threshold for considering empiric treatment; 105 (46% of those with a score above the threshold) were referred for treatment on the same day (implementation). Substantial variation in implementation was observed across clinics (Figure 1), with one clinic administering the PredicTB score to 86% (442 of 514) of all eligible patients and initiating same-day empiric treatment in 97% (38 of 39) patients with scores above the treatment threshold, contrasting with 21% (50 of 234) and 17% (3 of 18) in another clinic. For the one that used a cutoff of ⩾6 (vs. ⩾7 for all other clinics), 5 (12.5%) of 40 patients with a score of 6 were started on same-day treatment versus 14 (34%) of 41 with a score of 7 or higher.
Table 1.
Characteristics of people with presumptive tuberculosis across 11 primary health clinics in periurban Uganda before and during implementation of clinical risk score
| Intervention |
Standard-of-Care Comparison |
|||
|---|---|---|---|---|
| Preimplementation | Implementation Period | Preimplementation | Implementation Period | |
| (n = 4,295) | (n = 1,826) | (n = 2,211) | (n = 1,506) | |
| Age, yr, median (interquartile range) | 36 (27–47) | 37 (27–49) | 35 (27–46) | 35 (27–47) |
| Male sex | 1,775 (41%) | 749 (41%) | 1,016 (46%) | 671 (45%) |
| HIV status | ||||
| Positive | 1,788 (42%) | 857 (47%) | 964 (44%) | 684 (45%) |
| Negative | 2,064 (48%) | 760 (42%) | 996 (45%) | 741 (49%) |
| Unknown/missing | 425 (10%) | 209 (11%) | 250 (11%) | 81 (5%) |
| Sputum specimen submitted | 2,401 (56%) | 1,348 (74%) | 1,646 (74%) | 1,007 (67%) |
| Available clinical test results | ||||
| Xpert MTB/RIF* | 2,345 (55%) | 1,191 (65%) | 1,240 (56%) | 995 (66%) |
| Sputum smear microscopy* | 1,321 (31%) | 518 (28%) | 1,186 (54%) | 475 (32%) |
| Urine lipoarabinomannan* | 164 (4%) | 113 (6%) | 78 (4%) | 84 (6%) |
| Chest radiograph* | 46 (1%) | 45 (2%) | 139 (6%) | 70 (5%) |
| Mycobacterial culture† | NA | 61 (3%) | NA | NA |
| Microbiologically confirmed TB‡ | 354 (8%) | 173 (9%) | 205 (9%) | 138 (9%) |
Definition of abbreviations: HIV = human immunodeficiency virus; TB = tuberculosis; Xpert MTB/RIF = rapid molecular tool.
Percentages are reported out of the number of people with presumptive TB (not the number who submitted sputum).
Culture was not routinely performed but was performed through this study for all 61 individuals in the intervention arm who started empiric treatment but had negative results on other clinical tests.
Microbiological confirmation represents a positive test result on sputum Xpert, sputum smear microscopy, urine lipoarabinomannan, or sputum mycobacterial culture.
Figure 1.
Implementation of a clinical risk score to inform empiric tuberculosis (TB) treatment initiation in six Ugandan clinics. Shown is the cascade of implementation for a simple clinical risk score for TB (“PredicTB”) across six primary health clinics in periurban Uganda. Clinics C and D obtained access for on-site Xpert testing during the study period (after 54 and 228 d, respectively) and were replaced by clinics E and F for the remainder of the study period. Bars show the absolute number of individuals with symptoms (far left, black), individuals for whom the PredicTB score was used (second from left, dark gray), individuals whose PredicTB score was above the clinic-level threshold for considering empiric treatment (second from right, light gray), and individuals with scores above the threshold who started treatment on the same day (far right, white). Numbers initiating same-day treatment in this figure do not match those in Table 2, because this figure includes only those individuals who had a score assigned and whose score was above the clinic-level threshold for empiric treatment.
In the preintervention period among intervention clinics, 594 (83%) of 720 people diagnosed with TB (either clinically or microbiologically) started treatment within 7 days versus 264 (84%) of 316 after implementation (pre–post difference, 1%; 95% CI, −6%, 8%). In comparison clinics, 7-day treatment initiation changed from 312 (86%) of 363 preintervention to 153 (73%) of 211 postintervention (pre–post difference, −13%; 95% CI, −22%, −5%), for a difference-in-differences estimate of 14 percentage points (95% CI, 10%, 19%). Similar improvement among intervention clinics compared with comparison clinics was seen in the actual time to treatment initiation (decrease in time to initiation, 2.6 d [95% CI, −4.9, 0.3] in intervention vs. increased delay, 2.7 d [95% CI, 0.6, 4.7] in comparison clinics) (Table 2). Individual treatment outcomes did not change from preimplementation to postimplementation when comparing the intervention clinics with the comparison clinics overall; however, this overall trend reflected improvement in some clinics balanced by declines in others (Table E2). Similar results were seen when restricted to people with microbiologically confirmed TB (Table E3).
Table 2.
Treatment initiation and outcomes for individuals diagnosed with tuberculosis (either clinically or microbiologically) before and during implementation of clinical risk score
| Intervention |
Standard of Care |
Difference in Differences (Intervention – Standard of Care) | |||||
|---|---|---|---|---|---|---|---|
| Preimplementation (n = 740; 161 clinic-months) | Implementation Period (n = 323; 53 clinic-months) | Pre–Post Difference | Preimplementation (n = 382; 126 clinic-months) | Implementation Period (n = 219; 50 clinic-months) | Pre–Post Difference | ||
| Number of individuals with valid treatment and diagnosis date | 720 | 316 | — | 363 | 211 | — | — |
| Time to treatment initiation (mean, 95% CI) | 5.3 d (4.0, 6.7) | 3.7 d (2.7, 4.8) | −2.6 d (−4.9, 0.3) | 4.0 d (2.9, 5.1) | 6.8 d (4.8, 8.7) | 2.7 d (0.6, 4.7) | −5.4 d (−2.0, −8.6) |
| Same-day treatment initiation | 453 (63%) | 208 (66%) | 3% (−9, 16) | 233 (65%) | 116 (55%) | −10% (−14, −7) | 13% (8, 18) |
| Treatment initiation within 7 d | 594 (83%) | 264 (84%) | 1% (−6, 8)* | 312 (86%) | 153 (73%) | −13% (−22, −5) | 14% (10, 19) |
| Treatment outcome† | |||||||
| Treatment completion | 604 (84%) | 263 (83%) | −1% (−6, 5) | 296 (82%) | 162 (77%) | −5% (−12, 2%) | 4% (2, 6) |
| Lost to follow-up | 15 (2%) | 8 (3%) | 1% (−3, 2) | 12 (3%) | 9 (4%) | 1% (−5, 3) | 0% (−1, 0) |
| Failed | 1 (0%) | 0 (0%) | 0% | 0 (0%) | 1 (0%) | 0% | 0% |
| Died | 37 (5%) | 12 (4%) | −1% (−4, 1) | 31 (9%) | 17 (8%) | −1% (−6. 5) | −1% (−2, 0) |
Definition of abbreviation: CI = confidence interval.
Primary outcome.
Treatment outcome was not recorded or evaluated for 20, 7, 20, and 8 individuals, respectively, because transfer to nonstudy clinics or missing data from routine registers.
Patient scores on the PredicTB tool were correlated with TB risk (Table 3), with a cutoff of ⩾7 showing a sensitivity of 45%, specificity of 88%, and positive predictive value of 25% relative to microbiological results (Table E4). Of the 17 providers who evaluated patients for TB in the intervention clinics, adoption and acceptability of the PredicTB score were high (adoption; Figure 2). The primary reasons for not using the PredicTB score were that patients had mild symptoms or did not appear ill, but only 3 (18%) of 17 providers cited each of these concerns. The primary reason for not empirically treating patients with PredicTB scores above the clinic-level threshold was a desire to wait for Xpert results (reported by 8 of 17 providers; 47%). Direct observation indicated that calculating the risk score required an average of 2.14 minutes, which was short enough not to interrupt routine care.
Table 3.
Diagnostic performance of PredicTB clinical risk score in periurban Ugandan clinics
| Score | Number of Individuals | Number (%) with Confirmed TB* |
Model Estimate of Test Positivity† |
Sensitivity (%) | Specificity (%) | |
|---|---|---|---|---|---|---|
| 8% overall prevalence | 5% overall prevalence | 10% overall prevalence | ||||
| 1 | 4 | 0 (0%) | <1% | 1% | 100% (132/132) | 0% (0/1,074) |
| 2 | 62 | 1 (2%) | 1% | 2% | 100% (132/132) | 0% (4/1,074) |
| 3 | 207 | 11 (5%) | 2% | 4% | 99% (131/132) | 6% (65/1,074) |
| 4 | 298 | 10 (3%) | 5% | 9% | 91% (120/132) | 24% (261/1,074) |
| 5 | 267 | 24 (9%) | 9% | 17% | 83% (110/132) | 51% (549/1,074) |
| 6‡ | 176 | 27 (16%) | 22% | 37% | 65% (86/132) | 74% (792/1,074) |
| 7‡ | 106 | 27 (25%) | 26% | 43% | 45% (59/132) | 88% (941/1,074) |
| 8+ | 86 | 32 (37%) | 30% | 48% | 24% (32/132) | 95% (1,020/1,074) |
Definition of abbreviation: TB = tuberculosis.
Confirmed TB was defined as a positive result on sputum Xpert, sputum smear microscopy, urine lipoarabinomannan, or sputum mycobacterial culture.
Reference 17.
One clinic used a score of 6; the rest of the clinics used a score of 7 as the cutoff for same-day treatment initiation. Estimations were made on the basis of the corresponding cutoff for each clinic.
Figure 2.
Barriers to use and implementation of a simple clinical risk score for tuberculosis (TB) in periurban Ugandan clinics. Shown are responses from 17 providers who reported diagnosing or treating people with presumptive TB in six periurban Ugandan clinics where a simple clinical risk score (“PredicTB”) was implemented. Bars represent responses on a five-level Likert scale from “strongly agree” (white) to “strongly disagree” (black). The absolute number of responses from 17 providers is labeled in the bars. (A) Responses to the question, “I did not use the PredicTB score because….” (B) Responses to the question, “I did not treat patients whose score was above the PredicTB threshold because….” HIV = human immunodeficiency virus.
The average monthly percentage of individuals treated empirically (rather than microbiologically) during the implementation period was 48% (95% CI, 34%, 82%; maintenance) (Figure 3). The proportion of individuals who were treated empirically did not change substantially over the implementation period (Figure E2) or from the preimplementation to the implementation period (Figure E3). Similarly, the percentage of individuals with microbiologically confirmed TB who were treated within 7 days did not change substantially over time in the intervention clinics and was slightly lower in the intervention arm (52%) than in the standard-of-care arm (59%).
Figure 3.
Empiric treatment over time in the clinics implementing the PredicTB clinical risk score in periurban Ugandan clinics. Shown is the proportion of individuals empirically treated each month. The black line represents the mean proportion from all intervention clinics at each month. Overall, 48% of individuals were treated empirically (minimum, 34%; maximum, 67%). The blue line represents the LOESS (locally estimated scatterplot smoothing) smoothed trend, and the dark gray area shows the corresponding 95% confidence interval.
Discussion
In this pragmatic evaluation of six primary health clinics in periurban Uganda assigned to implement a simple clinical risk score to guide empiric treatment decisions for TB, 7-day treatment initiation for TB did not increase, comparing the preimplementation period with the implementation period. However, when compared with five comparison clinics that did not implement the risk score, clinics using the risk score had higher 7-day treatment initiation during the intervention period. This is primarily because 7-day treatment initiation declined over time in comparison clinics. Given that the comparison clinics also experienced an increased proportion of laboratory-confirmed TB diagnoses after receiving training on routine TB care and strengthened record-keeping activities, the training may have increased levels of laboratory-based TB confirmation in these clinics but may also have reduced empiric treatment among patients at high risk. The risk score was readily implemented and highly acceptable to providers but did not achieve a positive predictive value above 40%, even at a cutoff that included only 8% of the population. Although efforts to expand access to rapid microbiological testing for TB continue, further research is warranted into the use of simple tools to guide empiric treatment decisions in settings where same-day testing remains unavailable.
We observed substantial variation in reach and implementation of this clinical risk score, reflecting the pragmatic nature of our evaluation (21). Specifically, reach (the percentage of patients with symptoms for whom the score was calculated) varied from 21% to 86%, and implementation (the percentage of patients with scores above the clinic-level threshold for whom empiric treatment was initiated) varied from 17% to 97%. The clinic with the highest performance on these indicators was part of the initial implementation and had a high volume of patients with TB symptoms. By contrast, the clinics with the lowest performance had lower patient volumes and often had less time to normalize implementation, because they either received on-site Xpert machines and were thus discontinued or were brought on as “replacement” clinics. These clinics also tended to diffuse responsibility for implementing the PredicTB score across multiple providers (cough monitors, TB focal providers, nurses, and clinicians), whereas higher-performing clinics tended to centralize this responsibility in the hands of one or two individuals.
Even though reach and implementation of the PredicTB score were suboptimal at the clinic level, providers generally voiced strong support for the score as a decision-making tool. Furthermore, the primary reasons for not calculating the score or not acting on scores above the threshold reflected real-world clinical decision making—patients with mild symptoms or not appearing sufficiently ill to merit empiric treatment, wanting to wait for Xpert results—rather than distrust of the score, gaps in knowledge, or inappropriate use of the score (22). These findings illustrate the importance of further research into simple tools that can enhance clinical decision making in resource-limited settings, even if the present score did not increase the number of patients with confirmed TB who start treatment within 1 week of presentation.
Compared with the initial validation of the PredicTB score (17), data from this pragmatic evaluation showed somewhat lower discrimination. For example, a score of 7 (the most common cutoff used) was associated with only a 25% probability of confirmed TB in this setting of 8% overall test positivity versus 26% (assuming 5% positivity) to 37% (assuming 43% positivity) in the initial validation set. This reduced performance reflects, in part, a narrower distribution of scores in the present study (e.g., more than 60% of individuals evaluated for TB had a score higher than 7), but it also indicates that higher scores may be more appropriate thresholds for initiating empiric treatment in the field. These findings illustrate the challenges of transporting data on accuracy of clinical prediction rules—even those validated using data in high-burden settings—into the realities of programmatic management in highly resource-limited clinics. Specific challenges include 1) different spectra of symptoms in different clinical environments; 2) imperfect reference-standard data in routine care; 3) different valuations of false-positive and false-negative results in clinical practice; and 4) the complex realities of clinical decision making that also take into account patient presentation, local clinical experience, and variable trust in existing systems. These data also demonstrate that simple clinical risk scores should not be used as a replacement for expanding same-day microbiological testing.
Limitations
Our results should be interpreted in light of certain limitations. First, data in this pragmatic study were abstracted from each clinic’s routine registers, resulting in data of heterogeneous quality. Second, the pragmatic nature of our study may have resulted in an inability to detect a clinically meaningful difference in outcomes, even if the score improved care for individual patients. Third, the TB culture testing process was delayed for many patients, which may have led to false-negative results on culture. Fourth, we had a small sample size during the implementation phase, especially in the clinics that received on-site Xpert capacity during the implementation period and could therefore no longer participate in the study. Implementation of the risk score appeared to be less robust in clinics that were used to replace these sites; our results may therefore be biased toward the null. Fifth, heterogeneity in implementation may reflect unmeasured facility-level changes over time. Although we asked about major changes during our quarterly site visits, we cannot exclude the possibility of bias resulting from more gradual changes that did not rise to the level of clinic staff informing us of their effects. Although our results should be interpreted in light of these limitations, our study represents the real-world environment where a clinical risk score tool would most likely be used and as such provides important information regarding the implementation of clinical risk scores in settings with limited access to same-day TB diagnostics.
Conclusions
In conclusion, in this pragmatic evaluation of 11 primary health clinics in Uganda, a simple clinical risk score did not increase the proportion of people with confirmed TB who initiated treatment within 7 days. Implementation of this score was variable, however, and acceptability was high. Further research is therefore warranted into the most appropriate use of simple tools to improve clinical decision making for TB treatment in settings where same-day microbiological testing is not currently available.
Acknowledgments
Acknowledgment
The authors are grateful to the healthcare providers and patients at the study clinics for their time and participation in this study.
Footnotes
Supported by the National Institutes of Health (R21AI161301 [D.W.D.], K01HL165095 [Y.B.] and T32AI007291 [J.S.]).
Author Contributions: Y.B.: Study design, funding acquisition, data analysis, data interpretation, and manuscript draft. M.M.: Data acquisition, data interpretation, and manuscript review. A.T.: Data acquisition, data interpretation, and manuscript review. M.L.: Data acquisition, data interpretation, and manuscript review. V.N.: Data acquisition, data interpretation, and manuscript review. A.S.: Data acquisition, data interpretation, and manuscript review. K.O.R.: Data analysis, data interpretation, and manuscript review. J.S.: Data analysis, data interpretation, and manuscript review. A. Kityamuwesi: Data acquisition, data interpretation, and manuscript review. A. Katamba: Study concept, funding acquisition, data interpretation, and manuscript review. D.W.D.: Study concept, study design, funding acquisition, data interpretation, and manuscript review.
The data that support the findings of the study will be deposited at the Johns Hopkins Research Data Repository. Restrictions may apply to the availability of these data.
This article has a data supplement, which is accessible at the Supplements tab.
Artificial Intelligence Disclaimer: No artificial intelligence tools were used in writing this manuscript.
Author disclosures are available with the text of this article at www.atsjournals.org.
References
- 1.World Health Organization. Geneva, Switzerland: WHO; 2023. Global tuberculosis report 2023. [Google Scholar]
- 2. Teo AKJ, Singh SR, Prem K, Hsu LY, Yi S. Duration and determinants of delayed tuberculosis diagnosis and treatment in high-burden countries: a mixed-methods systematic review and meta-analysis. Respir Res . 2021;22:251. doi: 10.1186/s12931-021-01841-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Brown S, Leavy JE, Jancey J. Implementation of GeneXpert for TB testing in low- and middle-income countries: a systematic review. Glob Health Sci Pract . 2021;9:698–710. doi: 10.9745/GHSP-D-21-00121. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Shaozae CK, Das D, Kumar M. Diagnosis of tuberculosis in low-resource settings: overcoming challenges within laboratory practice. EMJ . 2023 doi: 10.33590/emj/10302558. [DOI] [Google Scholar]
- 5. Reid SJ, Mash R, Downing RV, Moosa S. Perspectives on key principles of generalist medical practice in public service in sub-Saharan Africa: a qualitative study. BMC Fam Pract . 2011;12:67. doi: 10.1186/1471-2296-12-67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Theron G, Zijenah L, Chanda D, Clowes P, Rachow A, Lesosky M. et al. TB-NEAT Team. Feasibility, accuracy, and clinical effect of point-of-care Xpert MTB/RIF testing for tuberculosis in primary-care settings in Africa: a multicentre, randomised, controlled trial. Lancet . 2014;383:424–435. doi: 10.1016/S0140-6736(13)62073-5. [DOI] [PubMed] [Google Scholar]
- 7. Cattamanchi A, Reza TF, Nalugwa T, Adams K, Nantale M, Oyuku D. et al. Multicomponent strategy with decentralized molecular testing for tuberculosis. N Engl J Med . 2021;385:2441–2450. doi: 10.1056/NEJMoa2105470. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Davis JL, Dowdy DW, den Boon S, Walter ND, Katamba A, Cattamanchi A. Test and treat: a new standard for smear-positive tuberculosis. J Acquir Immune Defic Syndr . 2012;61:e6–e8. doi: 10.1097/QAI.0b013e3182614bc5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Lessells RJ, Cooke GS, McGrath N, Nicol MP, Newell M-L, Godfrey-Faussett P. Impact of point-of-care Xpert MTB/RIF on tuberculosis treatment initiation. A cluster-randomized trial. Am J Respir Crit Care Med . 2017;196:901–910. doi: 10.1164/rccm.201702-0278OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Nalugwa T, Shete PB, Nantale M, Farr K, Ojok C, Ochom E. et al. Challenges with scale-up of GeneXpert MTB/RIF in Uganda: a health systems perspective. BMC Health Serv Res . 2020;20:162. doi: 10.1186/s12913-020-4997-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Zawedde-Muyanja S, Katamba A, Cattamanchi A, Castelnuovo B, Manabe YC. Patient and health system factors associated with pretreatment loss to follow up among patients diagnosed with tuberculosis using Xpert MTB/RIF testing in Uganda. BMC Public Health . 2020;20:1855. doi: 10.1186/s12889-020-09955-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. MacPherson P, Houben RMGJ, Glynn JR, Corbett EL, Kranzer K. Pre-treatment loss to follow-up in tuberculosis patients in low- and lower-middle-income countries and high-burden countries: a systematic review and meta-analysis. Bull World Health Organ . 2014;92:126–138. doi: 10.2471/BLT.13.124800. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Zawedde-Muyanja S, Musaazi J, Manabe YC, Katamba A, Nankabirwa JI, Castelnuovo B. et al. Estimating the effect of pretreatment loss to follow up on TB associated mortality at public health facilities in Uganda. PLoS One . 2020;15:e0241611. doi: 10.1371/journal.pone.0241611. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Alipanah N, Jarlsberg L, Miller C, Linh NN, Falzon D, Jaramillo E. et al. Adherence interventions and outcomes of tuberculosis treatment: a systematic review and meta-analysis of trials and observational studies. PLoS Med . 2018;15:e1002595. doi: 10.1371/journal.pmed.1002595. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Jensen SB, Rudolf F, Wejse C. Utility of a clinical scoring system in prioritizing TB investigations – a systematic review. Expert Rev Anti Infect Ther . 2019;17:475–488. doi: 10.1080/14787210.2019.1625770. [DOI] [PubMed] [Google Scholar]
- 16. Aunsborg JW, Hønge BL, Jespersen S, Rudolf F, Medina C, Correira FG. et al. Bissau HIV Cohort Study Group. A clinical score has utility in tuberculosis case-finding among patients with HIV: a feasibility study from Bissau. Int J Infect Dis . 2020;92S:S78–S84. doi: 10.1016/j.ijid.2020.03.012. [DOI] [PubMed] [Google Scholar]
- 17. Baik Y, Rickman HM, Hanrahan CF, Mmolawa L, Kitonsa PJ, Sewelana T. et al. A clinical score for identifying active tuberculosis while awaiting microbiological results: development and validation of a multivariable prediction model in sub-Saharan Africa. PLoS Med . 2020;17:e1003420. doi: 10.1371/journal.pmed.1003420. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Loudon K, Treweek S, Sullivan F, Donnan P, Thorpe KE, Zwarenstein M. The PRECIS-2 tool: designing trials that are fit for purpose. BMJ . 2015;350:h2147. doi: 10.1136/bmj.h2147. [DOI] [PubMed] [Google Scholar]
- 19. Turyahabwe S, Bamuloba M, Mugenyi L, Amanya G, Byaruhanga R, Imoko JF. et al. Community tuberculosis screening, testing and care, Uganda. Bull World Health Organ . 2024;102:400–409. doi: 10.2471/BLT.23.290641. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Glasgow RE, Harden SM, Gaglio B, Rabin B, Smith ML, Porter GC. et al. RE-AIM planning and evaluation framework: adapting to new science and practice with a 20-year review. Front Public Health . 2019;7:64. doi: 10.3389/fpubh.2019.00064. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Giraudeau B, Caille A, Eldridge SM, Weijer C, Zwarenstein M, Taljaard M. Heterogeneity in pragmatic randomised trials: sources and management. BMC Med . 2022;20:372. doi: 10.1186/s12916-022-02569-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Atsma F, Elwyn G, Westert G. Understanding unwarranted variation in clinical practice: a focus on network effects, reflective medicine and learning health systems. Int J Qual Health Care . 2020;32:271–274. doi: 10.1093/intqhc/mzaa023. [DOI] [PMC free article] [PubMed] [Google Scholar]