# Lives saved by tuberculosis control and prospects for achieving the 2015 global target for reducing tuberculosis mortality

## Philippe Glaziou, Katherine Floyd, Eline L Korenromp, Charalambos Sismanidis, Ana L Bierrenbach, Brian G Williams, Rifat Atun & Mario Raviglione

Volume 89, Number 8, August 2011, 573-582

### Table 2. Estimated tuberculosis (TB) case fatality rates for notified and non-notified cases of TB, by human immunodeficiency virus (HIV) infection status, in high-income countries, eastern European countries and all other study countries, 1990–2009

Country group | HIV-negative |
HIV-positive |
||
---|---|---|---|---|

Normal prior distribution^{a} |
Posterior distribution^{b} |
Triangular distribution | ||

Mean (SE) | Mean (SE) | Mode (bounds) | ||

High-income countries (n = 60) |
||||

Not notified | 0.1 (0.01) | 0.12 (0.004) | 0.2 (0.05−0.3) | |

Notified | 0.05 (0.011) | 0.045 (0.001) | 0.1 (0.05−0.15) | |

Eastern Europe (n = 16) |
||||

Not notified | 0.35 (0.02) | 0.36 (0.007) | 0.4 (0.2−0.8) | |

Notified | 0.08 (0.02) | 0.1 (0.005) | 0.2 (0.1−0.4) | |

All other countries (n = 137) |
||||

Not notified | 0.42 (0.012) | 0.37 (0.01) | 0.4 (0.2−0.8) | |

Notified | 0.1 (0.012) | 0.04 (0.007) | 0.2 (0.1−0.4) |

SE, standard error.

^{a} In Bayesian statistical inference, a probability distribution that expresses one’s uncertainty about a quantity *p* before the data are taken into account.

^{b} In Bayesian statistical inference, the conditional probability distribution of a quantity *p* that is assigned after the data are taken into account.

Note: The parameters that determine the prior probability distributions and the choice of their shape (e.g. a normal versus a triangular distribution) were derived from pooled estimates from random-effects modelling of recent literature review results.^{5}^{,}^{10}^{,}^{11} Sources of data include information from demographic surveillance sites (DSS) and the treatment outcomes reported in DOTS cohorts and published in the annual series of WHO reports on global TB control.^{5} Importantly, the TB deaths reported from verbal autopsy in DSS are widely heterogeneous, which limits their value for estimating case fatality rates (CFRs).^{14} Cohort data on treatment outcomes are also unsuitable for estimating CFRs because: (i) patients classified as having died on TB treatment may have died from a cause other than TB; (ii) they refer only to patients with an evaluated treatment outcome; and (iii) they cover only deaths that occur during treatment. TB mortality data should ideally come from a national VR system in which records are cross-referenced with patient records from routine case surveillance.