Understanding the impact of disease control: TB epidemiology and the GFATM

Questions Routine OR SurveysResearch Size of NotificationsPrivate sect.PrevalenceActive CF the Cure rateVital Reg.Incidence problemDeaths Global Report S. AfricaReview S. Korea Direction NotificationsSent. sitesPrevalenceModelling of changeTr. outcomesIncidence MoroccoChennai S. KoreaWorld Reasons HIV; MDR;Risk factorsHIVDuration dis. for changeDiagnosticsSocial issuesMDR AccessPPM NairobiIndia S. Africa S. Africa Impact ofNotificationsDiagnosisPrevalenceDOTS Control Cure rateDrug supplyIncidenceIPT DeathsQual. control PeruCzech.IndiaCzech.

Size/Routine Highest TB rates per capita are in Africa 25 to to to 299 < to or more No Estimate per population The boundaries and names shown and the designations used on this map do not imply the expression of any opinion whatsoever on the part of the World Health Organization concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Dotted lines on maps represent approximate border lines for which there may not yet be full agreement. © WHO 2002

Size/OR Table 1. Estimating case detection rates from vital registration data. Red: Statistics South Africa. Blue: TB programme. Errors are fractional errors. 2001Error (%) TB deaths (k) All deaths (k)451.9 Unknown causes (k)56.2 Deaths due to HIV9.220 Completeness of reporting (%)90.05 Proportion of HIV deaths due to TB TB deaths (k)66.1 Proportion of cases HIV positive (%) CFR negative (%) CFR positive (%) CFR (%)37.0 Incidence (k)242.9 Notifications (k)192.9 CDR (%) TB deaths = Incidence  Case fatality rate

Prevalence surveys: National blue; sub-national red. Kolin Size/Surveys

ARTI  50 Civil servants South Korea Size/Research 1

Calculating sample sizes Size/Research 2

Direction/Routine

Direction/OR

Decline in TB, Alaska Incidence or death rate/10K/year source: Grzybowski Tubercle 1976 deaths 30%/yr t 1/2 = 2.3 yr cases 13%/yr t 1/2 = 5 yr Direction/Surveys

Direction/Research TB: elimination by 2050? Incidence/million/yr Projected incidence 100x bigger than elimination threshold in 2050 GP2: incidence falls 5-6%/yr Vaccines, drugs and risk factors?

Nairobi Reasons/Routine

Smoking and TB in Chennai 27k deaths and 16k controls, k TB deaths Gajalakshmi, V., Peto, R., et al. Smoking and mortality from tuberculosis and other diseases in India: retrospective study of adult male deaths and controls Lancet (2003) –515. Smoker Non-smokerOdds TB deaths Controls OR = 5.9 (4.5) F = % of all TB deaths among men in Chennai are attributable to smoking Reasons/OR

Reasons/Surveys Drug Resistance in Retreatment Patients (n = 1 508) SA rates weighted by Province

TB incidence among gold miners in SA DDR Reasons/Research 1

Smear positive disease in South African gold miners Incidence (%/yr) Prevalence (%) Dis.Duration (yr) HIV HIV Ratio Corbett et al Reasons/Research 2

Dynamics of pulmonary TB in Peru Pulmonary TB cases/100,000 DOTS 1990 PTB falling at 6%/yr case finding Impact/Routine

Impact/OR

South Korea Impact/Surveys

Decline in prevalence Kolin, Czechoslovakia Men: 20%  10%/yr Women: 26%  21% China Impact/Research

The last word… We must eradicate tuberculosis, and we must do it now, … All available resources must be used. Chemotherapy, computers, prophylaxis and prevention, case finding and kindness can be blended in a properly constructed epidemiological model which will tell us exactly where we are going and how fast. Davies, J.C.A. The Eradication of Tuberculosis in Rhodesia DPH, London School of Hygiene and Tropical Medicine, 1966

Estimating CDR 1.Assume that the 1997 estimate of CDR is correct so that we know the incidence in Assume that the trend in total notifications (all forms; DOTS non-DOTS) gives the trend in incidence. 3.Use this trend to work out the incidence of all forms in the year Assume that the SS+ incidence is 45% of the all-forms incidence for HIV- people; 35% for HIV+ people. 5.Calculate the SS+ incidence 6.Divide the notification rate in DOTS areas by the SS+ incidence to get the SS+ DOTS case detection rate.

The Design Effect Suppose we have k clusters with m people in each cluster so that the total sample size is n = km. D = 1 + (m  1)ρ where ρ, the intra-class (or intra-cluster) correlation coefficient, is the ratio of the between-cluster variance s b to the total variance so that where s w is the within-cluster variance. 1.All members of a cluster are identical, sw = 0, r = 1 and D = m. Effective sample size = n/m = k = the number of clusters. 2.Members of each cluster have no particular similarity sb = 0, r = 0, and D = 1. Effective sample size = n = number of people. For example, if each cluster contains m = 10 people and ρ =0.1 then 1 + (m-1)ρ = 1.9. m = 20 people and ρ = 0.2 then 1 + (m – 1)ρ = 4.8 Even a small value of the intra-cluster correlation coefficient, multiplied by the size of the cluster, could lead to a substantial increase in variance and reduction in effective sample size.