Estimates of TB incidence, prevalence and mortality Philippe Glaziou Cairo, October 2009
Outline Main sources of information Incidence From incidence to prevalence From incidence to mortality TB/HIV MDR-TB
Main sources of information Measurements Mortality (Vital Registration) Prevalence (Prevalence survey) Service coverage, (inventory, capture recapture) Trends Time series of notifications and programmatic data Expert opinion Size of non-notified TB population
Incidence
Estimating incidence (2007) Reference year: 1997 global consultation process. 64 country estimates updated later) Of proportion of cases being notified (expert opinion) N = notifications / year, r = case detection ratio From surveys of infection λ denotes the percent risk of TB infection, l is expressed per 100,000/year This approximation is very uncertain, it assumes that 1 ss(+) remains infectious for 2 years and transmits infection to 10 susceptible individuals every year
Incidence: other methods From disease prevalence surveys P = prevalence, d = weighted duration From mortality data (vital registration) m = deaths, f = case fatality rate Capture re-capture, 3 lists required, loglinear modelling to estimate cases not in any list, adjustment for dependencies
Incidence: Main source of information (reference year)
Trends in incidence may reflect trends in: Notifications (when there is no significant change in case finding effort) annual risk of infection (repeat tuberculin surveys) mortality (Brazil, South Africa) else, a flat trend (zero slope) when data are too difficult to interpret. E.g. Iraq, Pakistan
Trends in incidence (2007) Trends in incidence assumed to mirror: trends in prevalence trends in mortality trends in ARI (tuberculin surveys) flat trends in notifications Number of countries 12 3 18 18 161
Limitations In most countries, trends in incidence mirror trends in notifications -> constant case finding effort assumed Difficult to interpret trends in infection measured through repeat PPD surveys Trends in prevalence (repeat prev. surveys) and trends in incidence not necessarily parallel Difficult to incorporate several sources of data as the estimation process is constrained to one year of reference and one model for trends Uncertainty not documented
Upcoming changes Three main sources of data From measurements of prevalence, using simpler method From measurements of mortality Assessment of surveillance data using WHO Task Force framework and quantification of expert opinion (onion model) Improve assessment of trends Documentation of uncertainty
WHO Task Force Framework Are data reliable and complete? Do notifications reflect trends in incidence? Does VR data reflect changes in TB mortality TB notification data Complete, consistent Vital registration (VR) data Accurate and with high coverage Time-changes in notifications of cases and deaths Changes in case-finding, case definitions, ICD codes, coverage of surveillance systems, TB determinants IMPROVE surveillance system Evaluate trends and impact of TB control Do notifications include all incident cases? Does the VR system include all TB deaths? Apply "onion" model to identify where cases may be missing Inventory studies with existing or newly developed study registries Capture re-capture studies notifications incidence VR mortality data deaths UPDATE estimates of TB incidence and mortality If appropriate, CERTIFY TB surveillance data as a direct measure of TB incidence and mortality
Data reliability 1. completeness of notification data and other quality checks are all reports complete and compiled? 2. internal consistency is there more sub-national variability in notification rates than expected? is there more variability over time than expected? is laboratory diagnosis of documented quality? 3. external consistency are proportions and rates consistent with current knowledge on TB epidemiology?
Removing duplicates in Brazil (2005) dups new cases incidence rate change (%) Cured (%) change (%) before after before after before after 19,064 81,330 74,113 44.2 40.2-9.7 60.5 64.5 +6.7 Source: Bierrenbach A et al. Rev Saúde Pública 2007; 41(Supl. 1): 67-76
Misclassifications Are case definitions consistent with WHO definitions? Is laboratory performance satisfactory? Microscopy units with satisfactory EQA results (no major error AND less than 3 minor errors) > 90% of all units If culture used, positive growth in untreated smear positives > 90%
The Onion Model No access to health care Access to health facilities, but don't go Presenting to health facilities, but undiagnosed Diagnosed by public or private providers, but not notified Diagnosed by NTP or collaborating providers Recorded in notification data Undiagnosed cases Diagnosed but not notified cases Notified cases All TB cases
Documented guess of the size of the non-notified TB population Timor-Leste Thailand Sri Lanka Nepal Myanmar Maldives Indonesia Bhutan Bangladesh Timor-Leste Thailand Sri Lanka Nepal Myanmar Maldives Indonesia Bhutan Bangladesh do not go not diagnosed 0 10 20 30 40 50 60 70 no access ntp not notified 0 10 20 30 40 50 60 70 percent non ntp not notified total 0 10 20 30 40 50 60 70
From Incidence to Prevalence
Assumptions 1990-2007 P = I. d Duration d provided as point estimate for 12 categories of patients: Shorter in HIV+ DOTS < non-dots < untreated median DOTS = 1yr, non-dots = 1.8 yrs, untreated = 2yrs Smear neg mostly similar to smear pos Durations in HIV- vary between countries Proportion smear positive vary between regions
From incidence to prevalence All incident cases Estimation of %HIV+ presented separately HIV+ve HIV-ve smear-positive (45%) smear-negative (55%) smear-positive (35%) smear-negative (65%) DOTS nondots untreated DOTS nondots untreated notifications (DOTS/ nondots, ss+/other) DOTS nondots untreated DOTS nondots untreated
Limitations Need estimates in 12 case categories for Incidence Duration Inconsistent definitions for DOTS and non DOTS patients between countries No analysis of propagation of errors Very large number of uncertain quantities and parameters
Upcoming simplifications N/I = N/P / (N/P + 1/d) d denotes the average duration of disease in untreated TB [1], N: notifs, I: incidence, P: prevalence Duration: triangular distribution from 1 to 4 years, mode at 2 years (Hanoi) HIV+: ratio d+/d- ~N (0.31, 0.088) [2] [1] Borgdorff M. New measurable indicator for tuberculosis case detection. EID 2004; 10(9): 1523-1528 [2] Williams et al. Anti-retroviral therapy for the control of HIV-associated tuberculosis: modelling the potential effects in nine African countries. Submitted.
Global prevalence (all forms), old and new method, by WHO region AFR AMR EMR 600 200 400 500 150 300 400 300 100 200 Rate per 100,000 200 100 140 120 EUR 50 700 600 SEA 100 0 400 WPR rates notifs prev.best prev.old 100 80 500 400 300 60 300 200 40 20 200 100 100 0 1990 1995 2000 2005 1990 1995 2000 2005 1990 1995 2000 2005
Mortality Ideally: directly measured Vital Registration with high coverage and low rate of ill-specified causes of deaths Interim systems: sample VR, verbal autopsy studies Indirectly estimated: M = I. f i i Where i is a case category (notification and HIV status), f denotes case fatality
TBHIV
Measurements of TB/HIV incidence Empirical measurements from 64 countries (7 national surveys, 8 sentinel surveillance, 49 provider-initiated HIV testing data with > 50% of new TB cases tested for HIV) t = I+ / I ; proportion HIV-positive among incident TB; h = N+/ N, HIV in general population (UNAIDS); ρ, Incidence rate ratio
Prediction of TB/HIV incidence Linear model of logit-transformed t using logit-transformed h, slope constrained to 1 t denotes HIV in TB h denotes HIV in general population
Three estimates of incidence rate ratio
Upcoming change Account for ART: multiply the IRR by a best estimate of TB risk ratio on/off ART Rifabutin projections: RR ~ Triangular (0.15, 0.3, 0.55) Sources of uncertainty: IRR (HIV pos/neg) RR (on/off ART)
MDRTB
Multidrug Resistant TB Direct measurements in 113 countries (new cases), of which 102 countries also have measurements on retreatment cases with π = Pr(MDR new), c = incident cases (new or retreatment), r = reported retreatment episodes and n = notified new cases
MDR-TB (cont) In countries with no direct measurement, p predicted from logistic regression model with indirect predictors such as Gross National Income, retreatment ratio r/n; % HIV in TB Model predictions should be replaced with measurements from quality surveillance data
Very weak indirect estimates of MDR-TB Predictive model very weak, the predictors are only indirectly related to the outcome Input data from DRS often outdated Limited data on MDR in categories of retreatment cases Double counting (new patient re-registered as retreatment during the same year) Misclassifications (retx -> new)
During this workshop, we would like to review the quality of surveillance data update assessment of trends changes in case finding efforts changes in predictors of incidence (e.g. HIV, GDP, MDR?) update estimates of incidence