Constructing care cascades for active tuberculosis: A strategy for program monitoring and identifying gaps in quality of care

1Tuberculosis (TB) is the leading infectious cause of death globally. The World Health Organization (WHO) has highlighted “patient-centered care for all people with TB” as a central pillar of its post-2015 End TB strategy. The cascade of care (also called the continuum of care) is a useful model for evaluating patient retention across sequential stages of care required to achieve a successful outcome. The cascade helps to quantify gaps in care delivery, pointing to areas in which quality of care could be improved. Over the last decade, the HIV community has pioneered use of the cascade to evaluate care delivery in diverse populations. This model has subsequently been applied to other diseases. The care cascade is instrumental in tracking progress in the Joint United Nations Programme on HIV/AIDS (UNAIDS) 90-90-90 global strategy for HIV.

2Care cascades have only recently been used to evaluate TB care, although TB programs have a tradition of conducting cohort analyses and, more recently, of using patient pathways analyses to understand dropouts in care. In addition, WHO has outlined an onion model in which patient losses across different steps in care are visualized as a series of concentric circles, and this conceptual model informs our approach to the care cascade.

3The United Nations Secretary General’s Special Envoy on TB has called for more widespread use of care cascade analyses to help achieve the End TB strategy. In addition, National Strategic Plans for India and South Africa refer to closing gaps in the care cascade as a key component of their TB elimination strategies. We discuss approaches for estimating care cascade stages for individuals with active TB, describe uses of this model for targeting interventions to address gaps in care, and suggest areas for future research. We argue that the care cascade has two potential benefits: as an approach for quantifying TB outcomes and as a conceptual framework for examining the quality of health services across various stages of care.

4TB has a range of states, ranging from latent infection (in which bacilli lie dormant, controlled by the immune system) to subclinical disease (in which the patient has no symptoms but has microbiological or radiographic evidence of disease) to active disease (in which the patient has symptoms in addition to microbiological or radiographic findings). The current manuscript describes an approach for estimating the care cascade for active disease. We do not cover treatment of latent infection, which affects around one-quarter of the world’s population. Other articles provide guidance on constructing care cascades for TB subpopulations, including individuals with latent infection, children with active disease, individuals with HIV/TB coinfection, and household contacts of TB patients.

5In Fig 1 (panel A), we present a model for the TB care cascade, integrating the WHO onion model with elements of the HIV care cascade. Each cascade stage contains a step (i.e., the absolute number of individuals achieving a point in care) and a gap (i.e., the difference between steps, representing individuals with suboptimal outcomes). Recent studies in India and South Africa used this general approach to estimate national-level TB outcomes. These countries differ with regard to HIV prevalence, initial diagnostic tests used, and healthcare landscape (Table 1). The studies presented outcomes for 2013 despite being published in 2016 and 2017, respectively, because multidrug-resistant tuberculosis (MDR TB) outcomes take 3 years to be reported, given the long treatment duration.

6Outcomes and major gaps in each country cascade vary, highlighting different deficiencies in care (Figs 1 and 2 and Table 1). The South African program performed better in terms of individuals with TB in the population accessing a TB test (Gap 1) but achieved poorer treatment outcomes than India’s public sector. About 37% of all patient losses in the South African cascade consisted of individuals who experienced poor outcomes during therapy (Gap 4). In contrast, India’s TB program did a poorer job of case finding: 50% of all patient losses consisted of individuals with incident TB who did not access a TB test (Gap 1). For both countries, Gap 2 is the second largest contributor to patient losses. MDR TB cascade outcomes in both countries are very poor, with deficiencies at every stage.

7These two studies may provide insights into the situation in other countries with similar epidemiological contexts. In addition to focusing on other high–TB-burden countries, future cascade analyses should address high-risk populations in countries with a lower TB burden (e.g., immigrants in Europe) and countries with high MDR TB rates (e.g., former Soviet Bloc countries), which are epidemiological contexts not represented in the current literature.

8Members of our team contributed to the recent Indian care cascade analysis. We studied methods used in the South African cascade for further insights. Our prior research is relevant for estimating different cascade stages, including the number of individuals with TB in the population (NA, VC), the diagnostic gap (MP, RN), pretreatment loss to follow-up (PTLFU; RS, SS, VC, MP), and post-treatment disease recurrence (VC). Our team also includes an expert in the HIV care cascade (KM). Input was incorporated from members of our team by email and in-person discussions. Limitations of the analytical approach are described in the main manuscript and S1 Appendix.

9The approach for constructing a care cascade depends on the assessment’s primary goal, which may include the following: (1) large-scale evaluations for monitoring patient outcomes in national programs or (2) smaller-scale evaluations for identifying gaps in quality of care at the clinic, city, or district levels. Large evaluations may aim to achieve nationally representative estimates of patient outcomes, while smaller-scale evaluations may additionally collect data on process indicators (indicators of quality of care) to enable intervention development.

10Different approaches for estimating a care cascade have varied risks of bias. Recently published TB care cascades used data from different patient cohorts to estimate each stage—what we refer to as a routine data approach (S1 Appendix). This approach does not account for the patient population’s changing composition at each stage, introducing biases that may carry forward to subsequent stages. In a cohort-based approach, the same individuals are followed through each cascade stage, minimizing risk of bias and achieving higher internal consistency (S1 Appendix). This approach allows estimation of the transition time of patients across stages, which has implications for disease transmission. We encourage use of cohort-based approaches whenever possible, although this approach is more resource intensive. If representative sampling of health facilities is used, it may be feasible to estimate cascade outcomes with reasonable precision using moderate samples even for large countries such as India or China. For example, the Population-based HIV Impact Assessment Project uses primary data collection with representative sampling to estimate the HIV care cascade in several African countries.

11Another challenge in estimating a TB care cascade is that common diagnostic tests for active TB have relatively low (e.g., sputum microscopy) or higher but imperfect (e.g., Xpert MTB/RIF) sensitivity. Xpert MTB/RIF, a polymerase chain reaction (PCR)-based test, has 85% to 92% sensitivity for diagnosing pulmonary TB, including rifampin resistance, compared to 40% to 60% sensitivity for sputum microscopy, but most high-burden countries are still reliant on microscopy for detecting active TB. A considerable proportion of TB patients are diagnosed empirically, especially when sputum microscopy is the only test used. In contrast, HIV tests have very high sensitivity and specificity, allowing for accurate identification of HIV-infected individuals who should be followed through subsequent cascade stages. HIV viral load also provides a reliable biological marker of effective treatment. In contrast, the diverse forms of TB (e.g., pulmonary, extrapulmonary, drug resistant) and potential for disease recurrence pose unique challenges for estimating TB care cascades. We therefore recommend approaches for estimating each stage based on the primary diagnostic test used in a given setting and the specific form of TB.

12A challenge for estimating care cascades in many countries (e.g., India, Indonesia, and Pakistan) is that a large proportion of TB patients are managed in the private sector. Notification rates for these patients are low. They are often treated empirically, without undergoing bacteriological testing, and the quality of private sector care is poor in standardized patient studies.

13Given low private sector notification rates, representative sampling of private laboratories with TB testing capabilities could allow cohort-based tracking of patients starting from Step 2 (accessed a TB test). Audits of lab registers would identify bacteriologically diagnosed private sector patients who may not be notified to national programs. From Step 2, approaches for estimating cascade stages would be similar to those for the public sector; however, this approach does not account for private sector patients who are diagnosed empirically, without a TB test. As such, representative sampling of private clinics that deliver a high volume of TB care (e.g., qualified physicians participating in public–private mix projects) may also be necessary in settings with high rates of empirical treatment. Chart audits could identify patients at these clinics who are treated empirically, who could be followed for treatment outcomes and disease recurrence rates.

14We describe approaches for estimating the TB care cascade below and in S1 Appendix. In Table 2, we summarize approaches for measuring care cascade outcomes and suggest process indicators for each cascade gap that may reveal deficiencies in quality of care. Data for process indicators could be collected concurrently with cohort-based studies aiming to measure care cascade outcomes.

15Estimating the number of individuals with active TB in a population (Step 1) is valuable for national-level cascades because the number of individuals with TB who do not access a TB test (Gap 1) may be a large gap and may contribute considerably to TB transmission. The annual number of individuals with incident TB is the ideal metric for Step 1 because most programs report subsequent outcomes, such as the number of individuals who complete treatment, on a yearly basis.

16For most countries, incidence and prevalence estimates are routinely reported by WHO and are informed by country experts. Alternative estimates are available from the Institute for Health Metrics and Evaluation (IHME). However, WHO and IHME incidence estimates are partly extrapolated from notification data, which may have inaccuracies, especially where the private sector delivers a large proportion of TB care. When possible, we suggest validating WHO or IHME estimates against independent sources of information on TB burden, such as private sector TB drug sales. Mathematical models, incorporating data from population-based surveys of active or latent TB prevalence and mortality, may also be informative. Moreover, population-based prevalence surveys provide objective data on the number of individuals with active TB in the population, which can be used for longitudinal monitoring. Prevalence surveys may also provide information on Gap 1 process indicators (Tables 2 and 3), which can be used to monitor the population’s care-seeking behavior and the impact of TB public education programs on modifying this behavior.

17For Gap 1, individuals who die without accessing TB care are particularly concerning. Achieving accurate estimates of these individuals is challenging, given limitations in the accuracy of vital registration systems and medical certification of causes of death in many countries. Verbal autopsy may help refine TB mortality estimates in such settings.

18We define Stage 2 starting from when individuals with pulmonary TB reach a health facility and access TB tests (e.g., sputum microscopy, Xpert MTB/RIF) or when appropriate workup is initiated by a healthcare provider for individuals with extrapulmonary or pulmonary TB who might be diagnosed empirically. While estimating Stage 2 requires different methods for each form of TB, it provides valuable insights on gaps in care. For example, in the Indian and South African TB care cascades, about 310,000 (16% of those tested) and 69,000 (14% of those tested), respectively, were not successfully diagnosed or never received their diagnosis. Estimating Gap 2 is especially valuable for smear-negative, Xpert-negative, and drug-resistant TB, which are more difficult to diagnose. This gap may reveal patient losses from use of suboptimal diagnostic tests (e.g., sputum microscopy) or from poor adherence to algorithms for empirical diagnosis.

19Individuals with smear-positive TB evaluated with sputum microscopy are, by definition, likely to be diagnosed. A small proportion may be missed if they do not submit a second sputum sample (S1 Appendix), especially in locations where same-day microscopy has not been implemented. In settings using Xpert, because a single sputum sample is usually submitted, the number of individuals with Xpert-positive TB who access the test (Step 2) can be assumed to be the same as the number who get diagnosed with Xpert-positive TB (Step 3).

20In settings without more advanced diagnostic tests, individuals with smear-negative TB are diagnosed empirically. Most individuals who have negative sputum smears have conditions other than TB (e.g., bacterial pneumonia), making it challenging to estimate the number of true smear-negative TB patients evaluated at diagnostic facilities. Because the number of individuals with smear-positive TB in Step 2 can be more reliably estimated, the estimated ratio of individuals with smear-negative to smear-positive TB in a setting (a reflection of the sensitivity of sputum microscopy compared to a gold standard of culture) can be used to roughly estimate the number of true smear-negative TB patients who get evaluated at diagnostic facilities (S1 Appendix). Estimates of this ratio may be more relevant if based on high-quality local studies of the sensitivity of sputum microscopy in programmatic conditions. In settings using Xpert MTB/RIF as the primary test, a similar ratio method based on estimates of Xpert’s sensitivity can be used to estimate Step 2 for individuals with Xpert-negative TB (S1 Appendix).

21Estimating the number of true extrapulmonary TB patients who access appropriate workup is also challenging because clinical presentation and sensitivity of diagnostic tests vary depending on the site of disease. Studies that identify individuals with possible extrapulmonary TB who present to diagnostic facilities and follow them to determine the number who complete appropriate workup and get diagnosed may inform Step 2 and Gap 2 estimates. The number of MDR (or rifampin-resistant) TB patients reaching health facilities and accessing a TB test (Step 2) can be estimated using MDR TB rates in new and previously treated patients, which are available for most countries from WHO or national MDR TB prevalence surveys (S1 Appendix). Finally, estimating Stage 2 for children can be particularly challenging because of the low sensitivity of diagnostic tests in this population (S1 Appendix).

22PTLFU—loss of diagnosed patients prior to treatment registration—is a major point of attrition in TB programs. Most studies have examined this gap for smear-positive or drug-resistant TB patients. Few have examined this gap for smear-negative, Xpert-negative, or extrapulmonary TB patients. Future care cascade analyses should estimate this gap for all forms of TB.

23To measure PTLFU, many studies identify newly diagnosed TB patients in registers at diagnostic facilities and prospectively track them to see if they get registered at treatment centers, an approach which can also facilitate cohort-based estimates for remaining cascade stages (Table 2). While we agree with this approach, it can be challenging for a few reasons. First, in some settings, TB treatment initiation and official registration (or notification) do not happen concurrently. Patients may be lost to follow-up after starting therapy but before official treatment registration. Second, patients may get diagnosed in one location (e.g., a city) but start treatment elsewhere (e.g., a rural area), making follow-up difficult, especially since unique identification numbers are uncommon in many countries. Third, missing or illegible contact information often makes patient tracking difficult, especially in settings using paper records.

24Capturing patient information in electronic registration systems at diagnosis and treatment initiation may improve estimation of PTLFU. South Africa has introduced unique patient identification numbers along with a national electronic notification system to ensure patients attending different facilities are not counted multiple times. India is rolling out a similar system. Such systems may facilitate patient tracking across large geographic areas. Officially registering (i.e., notifying) patients at the time of diagnosis, as India is trying to do, may also improve estimation of PTLFU.

25Finally, interviewing patients at the time of treatment registration allows assessment of delays in care seeking, diagnosis, and treatment initiation, which are helpful process indicators (Table 2). Some interventions may impact PTLFU and time delays differently. For example, a South African study found that use of Xpert reduced treatment delays for rifampin-resistant TB patients without reducing PTLFU.

26Most national TB programs routinely report data on patients registered in treatment (Step 4) and who do not complete therapy (Gap 4), based on the WHO guidelines. Suboptimal Gap 4 outcomes consist of patients who are lost to follow-up, experience treatment failure (i.e., positive sputum smear or culture despite therapy), or die while on treatment.

27While estimating Stage 4 using aggregate numbers from TB programs may be helpful, we recommend using prospective patient-tracking approaches that allow for rigorous cohort-based care cascade estimates. For this approach, patients diagnosed with TB in Step 3 can be followed through Step 4 (treatment registration) and Step 5 (treatment completion) using clinical records (Table 2). This approach also allows elucidation of the time during treatment when most poor outcomes occur (e.g., intensive or continuation phase). Digital adherence technologies—including digital pillboxes and cell phone–based strategies—may also facilitate more accurate estimation of Stage 4 and timing of patient losses.

28Step 5 (treatment completion) can be assessed using treatment cards or registers in most national TB programs. However, estimating Step 6 (1-year recurrence-free survival) requires following patients after treatment completion. Post-treatment follow-up is not routine in most programs, though some national guidelines recommend such monitoring. Studies show high rates of post-treatment disease recurrence and death under programmatic conditions, highlighting the importance of evaluating these longer-term outcomes.

29Post-treatment disease recurrence is an indicator of quality of care, since recurrence may result from poor medication adherence during therapy or undiagnosed drug resistance. In settings where HIV coinfection is common, disease recurrence is often due to exogenous reinfection with a new TB strain. One-year TB recurrence-free survival may be a less useful outcome for the cascade in such settings, although high recurrence rates in these settings may indicate need for transmission control interventions. We recommend 12 months of post-treatment follow-up because most cases of TB relapse (91%) occur in this time period, based on a meta-analysis of clinical trials.

30To achieve accurate Gap 5 and Step 6 estimates, we recommend a cohort-based approach with prospective follow-up of patients who complete treatment because retrospective follow-up of patients who finish treatment may be compromised by higher loss to follow-up. In addition, Gap 5 can most accurately be estimated by collecting sputum samples for mycobacterial culture from symptomatic patients (for those who had pulmonary TB) or repeated clinical evaluation (for those who had extrapulmonary TB), which is not possible to do retrospectively. Patients who complete TB treatment should ideally be regularly reevaluated (e.g., every 3 months), for at least 1 year.

31The care cascade represents a valuable and feasible approach for monitoring TB programs. Unique challenges involved in constructing a TB care cascade include difficulties in estimating the number of individuals with active TB in the population, challenges in estimating the diagnostic gap (Gap 2) due to the suboptimal sensitivity of common diagnostic tests, and heterogeneity in approaches for estimating cascade stages for different forms of TB. In addition, the case-finding gap (Gap 1) includes individuals with TB who do not access TB tests for various reasons, including not having access to health facilities, not seeking care, and not being referred for TB testing after reaching a healthcare provider. Understanding which barrier contributes most to Gap 1 is an important undertaking that we have not covered in this manuscript. Some challenges involved in estimating the care cascade are not unique to TB—for example, use of written records and lack of unique identification numbers, which makes tracking patients across stages more difficult. Additionally, it is not easy to account for patients managed in the private sector in some countries, without conducting primary data collection.

32Despite these challenges, key cascade stages can be evaluated in most settings. While robust estimates of the number of individuals with active TB in the population may not always be available, cohort studies can be implemented in most settings starting from Stage 2 or 3 to estimate subsequent stages. Even without estimates of the number of individuals with active TB in the population, these research approaches can provide valuable insights for strengthening health systems by identifying gaps with the largest patient losses.

33There are limitations in the scope of what the care cascade model measures. For example, delays in care seeking, diagnosis, and treatment initiation may not be adequately captured; however, as described above, the care cascade also provides a framework for understanding how patients traverse stages in care, into which other process indicators can be embedded. If cohort-based approaches are used to measure the care cascade, data on some of these process indicators can be collected concurrently to gain additional insights into quality of care.

34Ideally, care cascade estimates would rely on robust survey data and longitudinal monitoring by health systems, including nationally representative TB prevalence and mortality data, electronic medical records for capturing notification and outcomes of private sector TB patients, and routine post-treatment follow-up to estimate TB recurrence. Countries currently have variable availability of these data and infrastructure.

35Patient outcomes may be improved by implementing interventions addressing the most concerning gaps, which may be related to case finding, diagnostic workup, linkage to treatment, retention in care, or medication adherence (to reduce TB recurrence) (Fig 3). Patient mobility (e.g., urban–rural travel) is a barrier for ensuring linkage to, and retention in, care in many settings. Written records often require healthcare workers to track patients through different paper registers for diagnosis, drug susceptibility testing, treatment initiation, and treatment monitoring, which may contribute to diagnostic and treatment delays.

36Robust electronic systems with unique identification numbers for tracking patients, linking them to care, and monitoring medication adherence in real time have the potential to improve gaps in the care cascade. Once patients are started on treatment, a holistic management approach, including provision of economic incentives and enablers, nutritional support, and care for comorbidities (e.g., substance use, depression), may also improve outcomes.

37Although important information can be obtained from routine programmatic data, dedicated cohort studies will yield the most accurate care cascade estimates, especially for stages such as recurrence-free survival, for which programs may not routinely collect data. If representative sampling is used, multisite cohort studies can produce accurate national-level care cascade estimates that could be used for longitudinal monitoring of outcomes.

38The care cascade has the potential to improve program monitoring and to inform targeting of interventions to improve case finding, diagnosis, linkage to treatment, retention in care, and recurrence-free survival for TB patients. Combined with other approaches, such as patient pathways analyses, the care cascade can provide critical information on quality of care to national TB programs. The model may refine estimates for the STOP TB Partnership’s 90-(90)-90 global targets, which include getting 90% of people with active TB on appropriate therapy, reaching at least 90% of key high-risk or underserved populations as part of this approach, and ensuring that 90% of those patients achieve treatment success by 2025 at the latest. By providing a systematic approach to evaluating care delivery, followed by corrective interventions, the care cascade may serve as an important tool for achieving the ambitious goal of reducing TB incidence by 90% by 2035, as envisioned by the End TB strategy.