Background Knowledge of the HIV status in patients with tuberculosis (TB) and vice versa is crucial for proper individual patient management, while knowledge of the prevalence of co-infection guides preventive and therapeutic strategies. The aim of the study was to assess if national disease databases on TB and HIV are adequate sources to provide this information.
Methods A two way capture–recapture analysis to assess the completeness of the registers, and to obtain the prevalence of TB-HIV co-infection in the Netherlands in the years 2002–2012.
Results HIV testing was performed in less than 50% of the patients with TB. Of the 932 TB-HIV infected patients, just 293 (31.4%) were registered in both registers. Under-reporting of TB-HIV co-infection ranged from 50% to 70% in the national TB register, and from 31% to 37% in the HIV database. Prevalence of TB-HIV co-infection in the Netherlands in 2012 was 7.1% (95% CI 6.0% to 8.3%), which was more than double of the prevalence estimated from the national TB database.
Conclusions TB-HIV co-infection is markedly under-reported in national disease databases. There is an urgent need for improved registration and preferably a routine data exchange between the two surveillance systems.
- PUBLIC HEALTH
Statistics from Altmetric.com
Data from national registrations or surveillance are used to assess prevalence of disease/health status and their time trends, serve as an early warning system for outbreaks, or contribute to evaluating the impact of interventions towards a priori set health goals. These data are a vital source of information for national and international policymakers. Despite these useful objectives, many surveillance systems provide an incomplete picture, examples of which can be found in almost any country, about both communicable and non-communicable diseases.1–4
The incidence of tuberculosis (TB) and HIV in the Netherlands is recorded in two separate national databases: The Netherlands Tuberculosis Register (NTR) and the database of the Stichting HIV Monitoring (HIV Monitoring Foundation SHM). There is no protocol for systematic data exchange between the two registers. Information on TB-HIV co-infection in the Netherlands is solely derived from the NTR.
HIV infection is the single strongest risk factor for the development of TB. The incidence of TB is between 20 and 36 times higher in HIV-positive patients compared to the HIV-negative population.5 The use of combination antiretroviral therapy (cART) reduces the risk for an HIV-infected individual to develop TB substantially.6–8 However, TB treatment outcomes in HIV co-infected patients are poorer, even in HIV-infected patients on cART, compared to patients with TB without HIV. A missed diagnosis of TB at the start of cART is associated with a high risk of developing TB-associated immune reconstitution inflammatory syndrome, and has been shown to have a negative effect on immune restoration in HIV-infected patients.9–11 Given these clinical consequences, knowledge of the HIV and TB status in newly diagnosed patients with TB and HIV is crucial for proper individual patient management. For public health, an adequate estimate of the prevalence of TB-HIV co-infection guides the development of preventive and therapeutic strategies, and underlines the need to implement these strategies.
Around 1200 new HIV infections are diagnosed in the Netherlands yearly. The majority of the HIV patients are men (80%), who acquired the infection through homosexual contact (72%). Almost 90% of patients with HIV in the Netherlands are on antiretroviral treatment. The epidemiology of TB in the Netherlands is largely driven by the incidence of the disease in the non-Dutch population. The estimated incidence of TB in the Netherlands decreased from 9.1 per 100 000 population in 2000 to 5.7 per 100 000 in 2012, while the contribution of the non-Dutch population increased from 73% to 80%, respectively.12 With an increasing number of patients with TB coming from countries with a noticeable HIV prevalence, it was expected that the prevalence of TB-HIV in the Netherlands would increase, but the NTR does not show such a trend.
The aim of the current study is to assess the completeness of reporting of TB-HIV co-infection from the NTR and SHM database, and to use the data on reporting to assess the prevalence of TB-HIV co-infection in the Netherlands. It is expected that the results of this study will provide an evidence base for stakeholders to design and implement appropriate TB-HIV guidelines, and provide resources that fit the underlying epidemiology in the Netherlands.
We used the NTR and SHM databases as the two sources for a capture–recapture analysis. The NTR records diagnosed TB patients as reported by all Municipal Public Health Services (MPHS) in the country. The SHM records clinical follow-up data from HIV patients as reported by physicians in any of the 27 designated HIV-treatment centres in the country. Designated data managers for both registries merged the data at the patient level using different combinations of the parameters date of birth (with or without actual day), sex, zip code (two or four digits) and country of birth.
After the merge, personal identifiers were removed from the data set before it was handed over for analysis, as required by both the steering committees overseeing the use of the registries. Given the type of analysis and the absence of identifiable information, there was no need for informed consent.
We identified TB-HIV co-infected patients in each of the registers who should be recorded in the other. For the NTR, these were notified patients with TB known to be HIV infected. For the SHM, these were HIV patients who had a TB diagnosis at the time of or after their HIV diagnosis. We allowed the TB diagnosis to be for a maximum of 3 months before the HIV diagnosis to capture those patients in whom TB was the presenting illness that initiated HIV testing. In both registers, the diagnosis of active TB is recorded using a similar algorithm that can include a diagnosis based on clinical symptoms only. TB-infection is not recorded in the SHM, making it impossible to be a misclassification of active TB. We restricted our analysis to the period 2002–2012. In 2002, the SHM implemented the current data collection procedures; complete data from the NTR were only available up to 2012 at the time of the data request.
For each year we calculated the total number of patients with TB-HIV, using the adjusted Petersen-Lincoln estimate.13 ,14 where denotes the total population including the estimate of those missed, n1 the number of patients with TB-HIV in NTR, n2 the number of patients with TB-HIV in SHM, and m2 the number of patients recorded in both the data sources. The associated variance of the estimate of the total population is
The results from the capture–recapture analysis were used to estimate the under-notification of TB-HIV co-infection in the two databases separately. The denominator for the prevalence of TB-HIV co-infection consisted of the notified TB-cases (regardless of HIV-status) in the NTR, assuming that all possible patients with TB in the Netherlands were recorded in the NTR. The variance of the number of patients with TB-HIV from the capture–recapture analysis was used to obtain the 95% CI for the prevalence estimate.
The NTR contained 12 707 patients with a TB diagnosis between 2002 and 2012. The number of notified patients with TB with an HIV-test result was low (4% in 2002), albeit improving over the years to 20% in 2006 (figure 1). At this point in time, the guidelines for the management of HIV infection in patients with TB formulated by the Committee on Practical Tuberculosis Control were implemented, stipulating that all identified patients with TB should be tested for HIV.15 This improved overall HIV testing in this population, which increased to 53% of notified patients with TB in 2012. HIV testing improved more for non-Dutch patients with TB compared to Dutch patients with TB (figure 1).
The combined database for the time period 2002–2012 contained 932 patients infected with TB-HIV , of whom 508 were recorded in the NTR, 717 in the SHM and 293 (31.4%) in both registers. Of those included in the SHM but not in the NTR, 19 (4.5%) were matched with an entry in the NTR but were not considered a TB-HIV case in the NTR due to missing (n=18) or negative (n=1) HIV test.
The results of the capture–recapture analysis are reported in table 1. The percentage under-reporting of HIV infection in notified patients with TB (in NTR) shows a modest improvement over the research period, and varied between 50% and 70%. Under-reporting of TB in HIV-infected patients (in SHM) slightly improved over the research period from 37% in 2002 to 31% in 2012. The percentage of under-reporting in the SHM was lower than the under-reporting in the NTR.
The overall estimated prevalence of TB-HIV co-infection in 2012 was 7.1% (95% CI 6.0% to 8.3%) of all notified patients with TB (figure 2). Using the data available in the NTR only, the TB-HIV prevalence was 3.2%. The anticipated increase in the prevalence of TB-HIV co-infection over time did not occur. The prevalence decreased during the study period in the capture–recapture analysis, which was less obvious in the NTR database.
We showed that both the NTR and the SHM database are an inadequate source for estimating the prevalence of TB-HIV co-infection in the Netherlands, given their under-reporting. By using a capture–recapture analysis, the prevalence of TB-HIV co-infection in notified patients with TB in 2012 was more than double the estimate derived from the NTR. Our data indicate that a better registration and/or a regular estimate by a capture–recapture analysis using both databases are necessary for a valid estimate of the prevalence of TB-HIV co-infection in the Netherlands.
An earlier study by Scholten et al,16 compared patients with TB-HIV notified by national TB and AIDS programmes from all 52 countries in the WHO Europe region with the WHO expected numbers. Although this study concluded that, in general, patients with TB-HIV were under-reported by both of these programmes, the data for the Netherlands showed a good match between expected and reported number of patients. This is not surprising since the NTR and the SHM are the direct sources for international reporting of TB-HIV co-infection. These WHO data for the Netherlands can, therefore, not be used for assessing the completeness of reporting until better registration in the databases is implemented.
The most likely reason for the under-reporting of TB-HIV co-infection in the NTR is the organisation of reporting, rather than patients being missed by the health system and denied care. The large majority of patients with TB in the Netherlands are diagnosed by pulmonary care physicians in the context of evaluation of symptoms.12 It is mandatory to report an established TB diagnosis to the MPHS, where all available information is uploaded to the national register (NTR). With the ability to register ‘HIV test unknown’ at initial registration and the actual HIV results on a page that is filled later during the TB treatment period, NTR procedures preclude consistent reporting of HIV status. When information is initially missing (eg, HIV status), there needs to be recurrent contact between the diagnosing physician and MPHS to update the register. There is no guidance on the frequency and intensity of this communication, giving room for MPHS-specific efforts to obtain these data. Together with a general reluctance to test patients with TB for HIV, this leads to a considerable amount of missing HIV data in the NTR. Such an MPHS-specific approach was also seen in the conduct of contact tracing for patients with TB, where differences in strategies led to differences in the implementation of contact tracing guidelines, negatively influencing its effectiveness.17 As the data made available for analysis were void of identifying information at the patient level as well as at the MPHS level, we were not able to explore this finding further.
In addition to under-reporting HIV status in the NTR, there is an obvious lack of HIV testing in patients with TB, despite national guidelines stipulating its need.15 This situation is not unique for the Netherlands. Data on HIV testing in patients with TB in European countries with a low prevalence of TB are often absent in the yearly WHO report.18 Where present, these vary widely with less than 1% of all patients with TB tested in Finland and up to 76% of patients tested in Malta. An overview of practises in HIV testing in patients with TB in 25 European priority countries with regards to TB-HIV co-infection estimated that just over 39% of all notified patients with TB were tested for HIV.19 In the low prevalence countries of USA, New Zealand, and Australia, testing for HIV in patients with TB is routinely reported and was above 70% in 2013, compared to just 41% in Canada.18
Studies have shown that uptake of HIV testing by patients with TB is generally high when the test is offered.20 ,21 Insight into provider-initiated reasons for not offering the HIV test in low prevalence countries is almost absent. A small study in a single TB centre in Rhode Island, USA, identified provider-associated barriers towards an opt out strategy for HIV testing in TB suspects and patients. These included unqualified staff, unwilling staff and time constraints. Perceived low prevalence of HIV infection, together with reservations towards costs were seen as logistical barriers.22 A study from the Netherlands, before the current guidelines on HIV testing for all identified patients with TB, reported an association between more frequent HIV testing and setting (4 largest cities vs rest of the country), a combination of pulmonary and extra-pulmonary TB, and the patient originating from Sub-Saharan Africa. Despite the fact that such a risk-based testing was the national policy at that time, HIV testing was carried out in a minority of eligible patients.23 When the frequency of HIV testing improved after 2006, it increased mainly the proportion of patients with TB with a negative test result. This indicates that the decision to test for HIV was indeed based on individual risk assessment.
Under-reporting of TB in the SHM is more difficult to explain. Clinical data of all HIV patients are recorded by designated data entry staff following an a priori approved protocol. Missing data are, therefore, most likely a result due to omissions in the medical files. The selection of patients in the analysis ensures that the results are not influenced by older episodes of TB which were not reported by the patient. Each patient in the NTR had, by definition, a new episode of TB. Selection of new patients with TB in the SHM was guaranteed by excluding patients with TB with a diagnosis of more than 3 months before the HIV diagnosis.
The initial hypothesis of an increased prevalence of TB-HIV co-infection in the Netherlands as a result of immigration needs to be refuted as, based on our data, the prevalence of co-infection decreased. This could be explained by assumptions on migration in earlier modelling studies not being met.24 Changing migration patterns and national laws with respect to migration will influence TB-HIV prevalence in the country, but these are hard to predict. In addition, earlier start of potent cART in the recent years prevents reactivation of latent tuberculosis infection and the development of active TB disease.
Validity of capture–recapture analysis depends on the assumption of (1) independence of sources, (2) equal ‘catchability’ in the sources and (3) a closed population.13 Registration in the NTR and SHM are completely separate processes, where the former is the responsibility of the MPHS, and the latter the responsibility of well-trained data collectors. Without formal procedures of data exchange between the two registers, the sources can be regarded as independent in which registration in one is not influenced by registration in the other.
Both sources are an intrinsic part of patient care by HIV specialist physicians and Public Health Officers. There is no evidence that there is a differential probability of being registered in the SHM database or in the NTR given certain patient characteristics. On the contrary, both registers are used for international reporting of disease burden as a result of the (near) completeness of the databases. In a study by van Hest et al25 on the completeness of recording of patients with TB, the NTR was used as one of the verification sources, given its (near) completeness. As such, the assumption of equal ‘catchability’ seems to have been generally met. Related to ‘catchability’ is the use of similar case definitions in the registers used for the analysis. The NTR and SHM register have similar algorithms to confirm TB diagnosis. As such, the registers should capture the same patients when TB-HIV co-infections occur.
The assumption of a closed population is not met, but it can be argued that the effect of this violation on the completeness of recording is minimal. Treatment for HIV is lifelong, while TB can occur at any stage of the disease. The risk of TB increases with increased duration after seroconversion and advancement of immunosuppression.26 ,27 Although HIV patients not on treatment can be monitored outside the network of HIV treatment facilities in the Netherlands (although it only happens sporadically), ART is only available in these centres. With a continued decrease in the number of patients presenting late (marked immunosuppression) for care at an HIV treatment centre,28 loss to follow-up of HIV patients in the critical period for developing TB would be minimal, resulting in an adequate recording of TB-HIV co-infection. In the TB population, treatment is short (6 months), with registration occurring early within this period. Even if the patients with TB leaves the country, his/her episode has been fully recorded.
A potential limitation of the study may be that missing matches could have occurred because of a restricted matching strategy. Although the current study used four different matching strategies, missed matches could have occurred when clerical errors existed in any of the databases. The merging of the databases was carried out by two highly experienced data managers with in-depth knowledge of their respective databases. We can, therefore, assume a high level of accuracy in the matching procedures.
Another limitation of the study is the absence of a common set variables other than TB and HIV in the two databases, precluding the opportunity to look in detail into the characteristics of missed patients with TB-HIV. However, such an approach was never the objective of the study. The information available was adequate to assess the magnitude of under-reporting and the estimated prevalence of TB-HIV co-infection in the Netherlands.
Countries with a low incidence of TB share crucial aspects with the Dutch system with respect to absence of HIV testing in TB patients, and an epidemiology driven by TB occurrences in the non-native population. Since it is to be expected that no surveillance system is able to capture all events, it is likely that similar capture–recapture analyses using national TB and HIV registers will identify under-reporting of TB-HIV prevalence.
We conclude that the estimated prevalence of TB-HIV infection in the Netherlands is markedly higher than routinely reported by the NTR. Neither the NTR nor the SHM is an adequate source to estimate TB-HIV prevalence in the Netherlands. Since an adequate estimate of this prevalence is needed to design appropriate public health interventions that are embraced by the necessary health professionals, there is an urgent need to review recording strategies in both processes. The current study showed that merging of the databases is possible and can provide valuable information on patients from these overlapping diseases.
What is already known on this subject
National registries and surveillance systems provide valuable information to national and international policymakers to assess prevalence, time trends and impact of health interventions. Surveillance data are often incomplete, precluding their optimal use in health-policy decisions.
The Netherlands Tuberculosis Register is the sole source of reporting of tuberculosis (TB)-HIV co-infection in the Netherlands, the prevalence of which has remained stable over the previous years despite changes in TB epidemiology in the country.
What this study adds
The study shows a marked under-reporting of TB-HIV co-infection in the Netherlands. This occurred both in the national HIV register and the tuberculosis register. The estimate of its prevalence is twice as high as reported from the Dutch tuberculosis register to international policy bodies.
The results are the first ever adequate prevalence of TB-HIV co-infection in the Netherlands using national disease registration databases. The study shows that merging and analysis of these databases is possible and contributes to better TB-HIV surveillance in the country.
There is an urgent need for systematic data sharing between the two national registers in the Netherlands.
Contributors FvL conceptualised the study, performed all analyses and wrote the first and all of the subsequent drafts of the manuscript. All other authors contributed significantly to data collection and the interpretation of the data, and to the reviewing of the manuscript. MH, NK, and HS performed the data-merging procedures.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.