Implementation and effect of intensified case finding on diagnosis of tuberculosis in a large urban HIV clinic in Uganda: a retrospective cohort study.

HIV is the most important risk factor for the development of tuberculosis (TB). People living with HIV are 21-34 times more likely to develop TB than the HIV-uninfected [1]. TB is a leading cause of death among HIV infected people with 22% of all deaths attributable to TB.

In 2004, collaborative activities to reduce the burden of TB-HIV co-infection were added to the Stop TB Strategy of the World Health Organization (WHO) [2]. These activities include provision of antiretroviral therapy (ART) and the “3Is”: intensified case finding (ICF), isoniazid preventive therapy (IPT) and infection control. ICF refers to regular screening of all people with or at high risk of HIV or, in congregate settings, for symptoms of TB disease followed by prompt diagnosis and treatment [3]. It is the “gatekeeper” of the “3Is” as it identifies eligible patients for IPT and aids infection control [3, 4]. It differs from passive case finding (PCF) which detects TB cases among symptomatic patients presenting for diagnosis and treatment of symptoms [5].

Uganda is one of the 22 WHO TB high-burden countries with an HIV prevalence of 6% [1, 6]. The case detection rate of 61% in 2010 falls short of the WHO target of 70%. In order to increase the number of HIV-positive patients diagnosed with TB, the Ugandan Ministry of Health (MoH) in October 2008 issued a mandate for all health clinics to start ICF. In collaboration with the National Tuberculosis and Leprosy Programmed (NTLP), it developed a screening form to be administered in all health care settings to aid in the detection of tuberculosis among high-risk groups (people living with HIV/AIDS, or contacts of known TB patients).

Three ways to implement ICF in our high-burden HIV clinic in Kampala were tested for feasibility in October 2009; (1) the individual screen method: every patient was screened upon entry to the waiting area; (2) the poster screen method: posters in English and Luganda (the most widely spoken local language) were placed in the waiting area and included educational bullet points about TB and TB-HIV co-infection, the ICF screening questions, and instructions to approach a designated peer supporter (an IDI patient who receives a reimbursement for provided services in the clinic) if a patient recognized any of these symptoms; and (3) the announcement screen method: a trained peer educator gave twice daily presentations in English and Luganda on TB and TB-HIV co-infection and the ICF screening questions in the waiting area, asking waiting patients to approach a trained peer supporter for screening if they harbored any of the symptoms. Each method was piloted on 2 clinic days and then compared. The first method necessitated 4 additional full-time screening staff who only managed to screen 78% of patients. Furthermore, the increased TB suspect workload was unsustainable due to the additional human resource requirements, although arguably the workload could decrease over time. The second method was not feasible as 23% of patients reported being illiterate. The third method identified a similar number of TB cases as the first method with less increase in workload. Therefore, we decided to implement ICF using the announcement screen method. In this study we set out to determine the associated effect on case detection of TB.

We found that implementation of ICF by the announcement screen method in a large HIV clinic led to a 76% increase in the number of TB suspects identified, with only a 30% increase in newly diagnosed TB cases. PCF during the post-ICF period continued to identify a similar number of TB suspects compared with the pre-ICF period, even though ICF was performed at an earlier stage of the clinic visit than PCF. Different subsets of patients were identified through the two parallel screening methods.

The reported yield of new TB cases identified by ICF varies greatly, depending on the HIV prevalence in the population at risk studied, the country-specific TB prevalence and the method of screening used (symptom-based or not). According to a recent meta-analysis of studies of ICF in HIV-infected people in resource-limited settings, it ranged from 2.2% in contact-tracing exercises to 8.2% in medical and antiretroviral clinics [11]. Factors associated with a higher yield of ICF included the screening of all patients regardless of symptoms, and ICF in countries with higher TB prevalence. Screening strategies also vary widely and usually involve a set of questions. Absence of current cough, night sweats, weight loss and fever was associated with a low probability of active TB disease [12], although this was dependent on the prevalence of TB in the screened population with the negative predictive value decreasing from 99.7% to 92.3% with an increase in TB prevalence from 1% to 20%.

The yield of our symptom-based screening method was lower: at 1.4% and 1.7% of the active attending patient population in the pre-ICF and post-ICF period, respectively. In Uganda, prevalence estimates of newly diagnosed TB identified by symptom-based ICF screening programmes vary between 7.2% in a rural cohort of ART initiators, 3.6% in a rural cohort of patients screened for ART eligibility and 1.4% among new patients enrolling in an HIV clinic in the same hospital complex [13–15]. Our study population represents the population of a “mature” HIV clinic, with the majority of patients on ART and therefore less likely to have unrecognized TB. Also, regular TB screening via PCF may have already reduced the burden of infectious patients and thus the nosocomial transmission of TB, leading to a reduced TB incidence in our patient population. Other explanations for the low yield found in our study include an inefficient method of screening with missing of TB suspects (possibly also due to stigma, evidenced by the large proportion of TB suspects identified by PCF after ICF had been implemented), or a substandard laboratory performance, although regular external quality control has not shown this to be the case. The contribution of missed smear-negative, culture-positive TB is unclear: we diagnosed smear-negative pulmonary TB among 12% of coughing TB suspects in both the pre- and post-ICF period, which was lower than an estimated 19% smear-negative, culture-positive TB diagnosed among IDI pulmonary TB suspects in an earlier study (unpublished data from [16]). The effect of routine use of TB culture or Xpert MTB/RIF in confirming these empirically treated TB cases should be investigated, as well as the cost-effectiveness of such strategies.

We attempted to implement ICF in a methodical and feasible way using a method based on announcements. The individual screening method would possibly have led to identification of more suspects, but at a greater cost, both monetary and human resource. We show that this method is feasible, but less effective than PCF. The population of TB suspects identified by this method of ICF was different from the one identified through ongoing PCF in the same period, as evidenced by their differing baseline characteristics and the higher TB suspect to case ratio in ICF versus PCF suspects. This could have been due to inadequate screening by the peer supporters, or to a higher likelihood of more “experienced” patients presenting themselves for screening. This correlates with our finding that only one-fifth of TB suspects who were newly enrolled in HIV care were identified through ICF. Conversely, considering that the suspects identified through PCF in both periods were relatively similar but differed from those of the suspects identified through ICF, ICF might have identified an additional subset of TB patients which otherwise would have remained undiagnosed and therefore infectious. Our results cannot confirm or exclude this possibility, however. Additionally, stigma attached to TB may result in the patient presenting to the clinician when being asked the screening questions rather than approaching the peer supporter in the waiting area. Incomprehension of the message may also be a cause; either due to a language barrier or because of the manner in which the message is delivered. Our ICF method was very dependent on how the person charged with giving the announcements did this, and how he/she emphasized the importance of knowing one’s “TB status”. The individual screening method may arguably have fewer of these problems. Lastly, the announcements could possibly have made PCF more effective, by sensitizing patients and making them more likely to volunteer their symptoms during their clinician visit.

Costs per TB case identified rose from 12 to 22 USD after implementation of ICF using the announcement screen method. The increase was mainly due to the higher suspect to case ratio and to the cost of the peer supporters performing the screening. These were relative costs, comparing costs pre- and post-ICF implementation. Absolute costs of screening programmes are highly dependent on the work-up of TB suspects and comparisons across settings and countries are therefore difficult to make. We lacked both cost and clinical inputs to do a formal cost-effectiveness analysis. Our analysis was simple in nature and therefore did not take staff time or treatment costs into account. The TB-HIV clinic staff including the peer supporters would only have helped out in the regular ART clinic on an ad-hoc basis. A formal cost-effectiveness analysis is needed to inform practice and should include the effect of ICF on TB transmission. The cost-effectiveness of screening of the whole population regardless of (the duration of) symptoms, as proposed by some [11, 17], would have to be established. This would not be feasible in our clinic without a comprehensive restructuring of the health care delivery at IDI and without a significant increase in resources for personnel and investigations.

Our findings raise the question how best to optimize ICF in a setting like ours. As the majority of TB cases were found among patients newly enrolled into HIV care, it seems advisable to screen that part of the patient population systematically, possibly using an individual screening method. For patients in care for a longer period of time, the optimal screening strategy is less obvious. As clinicians are less likely to think of TB in stable patients, there is a need for some form of additional screening, especially if it would sensitize the patient to report TB-symptoms during the formal encounter with the HIV clinician. Targeted announcements such as in our study might work for this category of patients. Optimal screening frequencies should be established, for example by comparing a strategy of screening at each clinic visit with 3-monthly, 6-monthly or even yearly TB screening.

Noteworthy was the high retention rate of TB suspects: 95% and 96% underwent investigations. This was substantially higher than in a similar screening programmed in Swaziland, where this was only 53% [18].

Limitations of our study were the retrospective study design and the use of routinely collected data with missing data as a result. The 50 excluded patients who were identified by ICF but did not qualify as a TB suspect on arrival at the clinic, added to the increase in workload after ICF implementation. Some of these were misidentified by the peer supporters, highlighting the need for continuous retraining of lay health care workers. Others purposefully reported TB symptoms at screening to ensure being seen by a clinician in the TB-HIV clinic. Interestingly, 32 patients were identified as TB suspects by the peer supporter but were never seen in the TB clinic, possibly due to long waiting times or to stigma associated with TB. The potential TB suspects for whom no data was available (n = 56) possibly belong to the same category. Both point towards a possible systems issue absorbing these patients in the TB-HIV clinic. Although we tried to highlight differences in study population, changes due to time-trends are a limitation of the before-after study design. We feel that the limited period of time in which the study was conducted was unlikely to have a major impact on the results.

Sputum samples were investigated by light microscopy using Ziehl-Neelsen staining in the pre-ICF period, while an in-house LED-based fluorescence microscopy was established at the start of the post-ICF period. However, we do not believe the 10% sensitivity difference to have influenced our case-detection in both periods greatly [19]. We made assumptions of the numbers of investigations ordered for the costing analysis; however, as the investigations did not differ hugely in price, we believe that using the actual numbers would not change our estimate markedly. We also assumed a flat cost of 5,000 UGX (2.16 USD) for sputum smear microscopy, while the in-house test implemented in the post-ICF period was cheaper (true costs were not available). Our costing analysis was therefore conservative.

Lastly, this method of ICF implementation would not have been possible in a large clinic without resources for the additional lay health care workers and investigations. The IDI has more available tests for diagnosis and better follow-up of the patients compared to most health care settings in sub-Saharan Africa. Also, the on-site fluorescence microscopy services could handle the additional sputum smears in a timely fashion. For ICF to be implemented countrywide in Uganda and in other resource limited settings, there would have to be a scale-up of resources in order to train health care and lay workers and to equip the laboratory systems to handle the extra workload.

Implementation of ICF in a large, “mature” HIV clinic using an announcement screen method yielded more TB suspects and cases, but substantially increased costs and was unable to capture the majority of TB suspects referred for diagnosis, potentially diluting an effect on infection control. The overall yield of TB cases in a mature HIV clinic was low, raising the question whether ICF is cost-effective and affordable in such a setting. Research into the optimization of ICF is needed, including targeted screening strategies for those recently enrolled into care.

Conceived and designed the experiments: SH, EN, FvL, EB, MS and YM. Analyzed the data: SH and FvL. Wrote the manuscript: SH, EN, FvL, EB, MS, AH, JL and YM. ICMJE criteria for authorship read and met: SH, EN, FvL, EB, MS, AH, JL and YM. Agree with the manuscript’s results and conclusions: SH, EN, FvL, EB, MS, AH, JL and YM. Wrote the first draft: SH and EN. All authors read and approved the final manuscript.