Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Eur Respir J. Author manuscript; available in PMC 2013 November 3.
Published in final edited form as:
PMCID: PMC3800490

Assessing spatial heterogeneity of MDR-TB in a high burden country


Multidrug-resistant tuberculosis (MDR-TB) is a major concern in countries of the former Soviet Union. The reported risk of resistance among TB cases in the Republic of Moldova is among the highest in the world. We aimed to produce high-resolution spatial maps of MDR-TB risk and burden in this setting.

We analyzed national TB surveillance data collected between 2007 and 2010 in Moldova. High drug susceptibility testing coverage and detailed location data permitted identification of sub-regional areas of higher MDR-TB risk. We investigated whether the distribution of cases with MDR-TB risk factors could explain this observed spatial variation in MDR-TB.

3,447 MDR-TB cases were notified during this period; 24% of new and 62% of previously treated patients had MDR-TB. Nationally, the estimated annual MDR-TB incidence was 54 cases/100,000 persons and >1,000 cases/100,000 persons within penitentiaries. We identified substantial geographic variation in MDR-TB burden and hotspots of MDR-TB. Locations with a higher percentage of previously incarcerated TB cases were at greater risk of being MDR-TB hotspots.

Spatial analyses revealed striking geographic heterogeneity of MDR-TB. Methods to identify locations of high MDR-TB risk and burden should allow for better resource allocation and more appropriate targeting of studies to understand local mechanisms driving resistance.

Keywords: antibiotic resistance, geographic heterogeneity, mapping, Moldova, spatial epidemiology


Intensification of global tuberculosis (TB) control efforts have contributed to declines in estimated incidence and mortality1. However, drug resistant TB threatens recent successes which are built upon the use of standardized drug regimens2. In 2009–2010, the highest ever levels of multidrug-resistant TB (MDR-TB, i.e. TB resistant to at least isoniazid and rifampicin and does not respond to first-line combination therapy3) were reported4,5. MDR-TB control is hampered by the challenges in both detecting resistant disease through drug susceptibility testing (DST) and access to quality assured treatment6. In 2010, of the 290,000 estimated MDR-TB cases among notified pulmonary TB cases globally, only 16% were diagnosed and initiated on appropriate treatment1.

Countries of the former Soviet Union (FSU) have reported percentages of TB cases with MDR-TB several times higher than countries elsewhere in the world1,4,7,8. The Republic of Moldova is a small country (area=33,846 km2, population=four million (2010)) that gained independence from the Soviet Union in 1991. Challenging economic conditions have resulted in substantial emigration, mainly to Europe, and currently 800,000 Moldovans are estimated to be abroad9. Like many areas of the FSU, Moldova has a serious MDR-TB problem: a 2006 nationwide TB drug-resistance survey found that 19.4% and 50.8% of new and previously treated TB cases respectively had MDR-TB8,10. Consequently, substantial investments were made in Moldova to improve TB diagnosis, surveillance and treatment7,11. Notable achievements include the expansion of DST coverage, which is mandated for all culture-positive cases (an exceptionally rare policy in high TB prevalence countries), and the development of an online data collection system that incorporates laboratory, clinical and demographic information.

Identifying where MDR-TB is concentrated geographically should help inform the rational allocation of resources and allow for targeted studies that aim to identify local drivers of emergence and spread of resistance. Here, we use programmatic data collected between 2007 and 2010 in Moldova to describe the risk and burden of MDR-TB both nationwide and locally. We also produce detailed maps to illustrate spatial variation in MDR-TB. Finally, we identify host-level risk factors of multidrug resistance among TB cases and investigate whether the spatial distribution of such risk factors can explain the observed geographic variation in MDR-TB.


Data sources

We used the Moldovan TB database of all notified TB cases diagnosed nationwide between January 2007 and December 2010 (23,152 cases). TB cases are diagnosed by sputum smear microscopy, culture, and/or abnormal radiography in the presence of symptoms. Detailed demographic and residential location data on each notified case are collected at the time of TB diagnosis and entered into the centralized online database. Laboratory results are added to this database when available and all data are verified by staff at the National Centre of Health Management. Data are also included from Transnistria, a semi-autonomous region in eastern Moldova which operates largely independently but contributes cases to the national TB database, Figure 1.

Figure 1
Map of the Republic of Moldova. Borders of the administrative regions are shown which highlight the 4 municipalities (in black) and 2 semi-autonomous regions (in grey) [note that Moldova is largely rural and the 4 municipalities shown are the main urban ...

Moldova has four laboratories performing culture and DST (including one national reference laboratory) which have passed external quality assurance conducted by the Supranational Reference Laboratory Network in Borstel, Germany8. DST was done on solid culture using the absolute concentration method10.

We obtained population estimates from the most recent census12 (2004). Moldova is divided, administratively, into 44 regions (“rayons”) and 1681 “localities” (villages or sections within a city). Thus, residential location data enabled us to pinpoint a case to a locality within a rayon.

Statistical analysis

Assessing diagnostic tools and MDR-TB burden

We calculated the percentage of notified pulmonary TB cases receiving smear microscopy, mycobacterial culture, and DST by year. MDR-TB cases were those with resistance to at least isoniazid and rifampicin. Non-MDR-TB cases were those who received DST and were susceptible to either isoniazid or rifampicin. Using data from TB cases with sufficient DST to classify as either MDR-TB or non-MDR-TB, we calculated the percentage of cases with MDR-TB separately for new TB patients (i.e. those who have either previously received treatment for less than one month in total or never) and previously treated TB patients (i.e. those who have previously received one month or more TB treatment)13.

Since TB cases without a culture-positive sample did not receive DST, MDR-TB notifications likely underreport the total MDR-TB burden. We used two different approaches for estimating the actual burden of MDR-TB. First, as a conservative lower bound, we estimated MDR-TB incidence assuming that only notified TB cases with positive smear and/or culture could have had MDR-TB. Second, we estimated the MDR-TB incidence assuming that all notified cases (i.e. cases with or without microbiological confirmation) could have had MDR-TB. Both of these approaches assume that cases without DST results were as likely to have had MDR-TB as those who were tested since DST was carried out for nearly all culture positive TB cases (and not targeted only to those presumed more likely to have MDR-TB).

We report use of diagnostic tools, incidence of notified new and previously treated MDR-TB cases, and percentage of TB cases with MDR-TB by rayon. When estimating MDR-TB incidence stratified by rayon, we used rayon-specific estimated percentages of MDR-TB, stratified by new or previously treated status, and rayon-specific notified TB incidences. We repeated these analyses for patients diagnosed within the penitentiary system.

Individual-level factors associated with MDR-TB among TB cases

We constructed two logistic regression models to identify individual-level (i.e. host-specific) variables that were significantly associated with the odds of having MDR-TB among new and previously treated TB cases. Factors which increase an individual’s odds of having MDR-TB among these two groups may differ since resistance among new cases reflects MDR-TB transmission whereas resistance among previously treated cases may result from transmitted resistance or resistance acquired during prior exposure to TB drugs.

A full model was constructed including all potential explanatory variables to obtain fully adjusted odds ratios. Some cases were missing data on some variables (Appendix Table 1) and hence inclusion of these variables would reduce the sample size used in the model. Therefore, we examined only those variables for which less than ten percent were missing data. A backwards elimination method was used to identify variables that were statistically significantly associated with MDR-TB diagnosis. In addition, any non-statistically significant variable which, on removal, altered other parameter estimates substantially (>10%) remained in the model to ensure full adjustment for confounding.(Appendix Methods).

Geographic heterogeneity of the incidence of MDR-TB

In addition to the analysis of MDR-TB burden by rayon described above, we constructed maps of MDR-TB burden. The percentage of TB cases with MDR-TB and notified and estimated MDR-TB incidences were aggregated by locality and plotted using latitude and longitude co-ordinates. Since these provided estimates at specific locations, we used inverse distance weighting14 to produce maps which allow visualization of smoothed estimates of incidence across the entire country. This method divides the country into cells of a pre-specified area and estimates the value of interest (i.e. MDR-TB incidence or percentage with MDR-TB) in each cell. This is done by calculating an average of the values from the nearest N points inversely weighting them by the distance of each point from the cell so that nearby points have a large influence and further points have less (for technical details see Appendix Methods). To test for local spatial clustering and outliers of high MDR-TB risk, we used Anselin Local Moran’s I15 (Appendix Methods).

We then assessed whether a concentration of cases with factors that increased the odds of having MDR-TB (identified by the methods described above) could explain the increased MDR-TB burden observed in some localities. We used two logistic regression models (for new and previously treated TB cases) to model the proportion of TB cases with DST that had MDR-TB in each locality. For each individual-level factor identified above, we created an explanatory variable for the proportion of all TB cases in the locality which had that factor. All variables were log-transformed before inclusion in the model16. We assessed the predictive ability of the model using receiver operating characteristic (ROC) curves where an Area Under the Curve (AUC) value represents the predictive probability of the model (AUC=0.5, none; AUC=0.5–0.7, poor; AUC=0.7–0.9, reasonable; AUC>0.9, very good)17 (Appendix Methods).

We did not make any adjustment for potential spatial variation in TB case detection rates. We made the assumptions that 1) the percentages of undetected TB cases that had MDR-TB were equivalent to those among detected TB cases and 2) there was no substantial spatial variation in TB case detection rates. Analyses were carried out using SAS version 9.2 and ArcMap version 10.0.


From 2007 to 2010, there were 23,152 notified TB cases in Moldova (Appendix Table 2). Ninety-eight percent of notified pulmonary TB cases had sputum specimens examined by microscopy and 92% received culture examination (Appendix Figure 1). Approximately half of all notified pulmonary TB cases were culture-positive and 94% of notified culture-positive TB cases received DST. Generally, the use of diagnostic tools in the penitentiary system was comparable with the civilian sector. However, in Transnistria only 59% of pulmonary TB cases received culture and only 73% of culture-positive cases received DST (Appendix Table 2).

Between 2007 and 2010, there were 3,447 notified MDR-TB cases accounting for 38% of TB cases that received DST for MDR-TB (23.5% of new cases and 61.5% of previously treated cases had MDR-TB, Table 1). Among new cases, younger cases were more likely to have MDR-TB (Figure 2). While the annual rate of notified MDR-TB/100,000 people was 20.9 cases, this statistic underreports the total burden since many TB cases were culture negative and therefore unable to receive a DST. Therefore, we estimated an annual MDR-TB incidence of as much as 54 cases/100,000 people, dependent on the method used to adjust for underreporting (Table 1).

Figure 2
Number of tuberculosis (TB) cases confirmed to have multidrug-resistant tuberculosis (MDR-TB) as a percentage of those with sufficient drug susceptibility testing to diagnose MDR-TB by age, 2007–2010. Pale grey bars indicate new TB cases, darker ...
Table 1
Burden of multidrug-resistant TB (MDR-TB) in Moldova, 2007–2010. The number of notified MDR-TB cases, percentage of TB cases with MDR-TB and the notified and estimated incidence of MDR-TB by case type. Results for the entire country and the subset ...

Within the penitentiary system, there were 1,689 TB cases (7.3% of all notified TB cases in Moldova). In prisons outside Transnistria, 37% of new and 83% of previously treated cases had MDR-TB (Table 1). In Transnistrian prisons, more than three-quarters of TB cases had MDR-TB. The estimated MDR-TB incidence in the entire penitentiary system may be greater than 1,000 annual cases/100,000 persons; more than 20 times that estimated for the rest of Moldova.

Individual-level factors associated with MDR-TB among TB cases

We identified several characteristics for new and previously treated TB cases that were independently associated with having MDR-TB (see Table 2 for full list). Several potential factors were not independently associated with MDR-TB in new or previously treated TB cases such as gender, smear status, education level and household size (Appendix Table 10).

Table 2
Individual-level risk factors (i.e. those that are associated with an increased disease risk for an individual) for multidrug-resistant tuberculosis (MDR-TB) diagnosis in new and previously treated tuberculosis (TB) cases. Cells left blank indicate that ...

Geographic heterogeneity in MDR-TB incidence

Both the percentage of tested TB cases with MDR-TB and the estimated MDR-TB incidence varied substantially between rayons (Appendix Figure 2). Among new TB cases, the percentage with MDR-TB ranged from 2.9% to 41.4% and among previously treated cases from 41.0% to 78.9%. The ranges in the annual notification and estimated annual incidence of MDR-TB were similarly wide (Appendix Figure 2).

Mapping of notified MDR-TB cases by locality revealed marked heterogeneity even within single rayons (Figures 3 and and4).4). Spatial clustering analysis highlighted several areas with statistically significant clustering of high MDR-TB risk and localities that had substantially higher MDR-TB risk than those around them (Figures 3(c) and 3(d)). Our locality-level regression showed that a higher risk of MDR-TB among both new and previously treated cases was most strongly associated with a local increase in the percentage of TB cases that had previously been in detention (Table 3). However, ROC analysis showed that the models had poor predictive probabilities (AUC=0.59 for each model) indicating that a substantial amount of the local heterogeneity in MDR-TB remained unexplained by individual-level characteristics that could be included in our models.

Figure 3
Maps of multidrug-resistant TB (MDR-TB) in the Republic of Moldova. Number of notified MDR-TB cases as a percentage of TB cases with drug susceptibility testing (DST) sufficient to diagnose MDR-TB among (a) new TB cases and (b) previously treated TB cases. ...
Figure 4
Annual notified MDR-TB incidence per 100,000 population among (a) new TB cases and (b) previously treated TB cases and annual estimated MDR-TB incidence per 100,000 population among (c) new TB cases and (d) previously treated cases. All maps were produced ...
Table 3
Locality-level risk factors for MDR-TB diagnosis in new and previously treated TB cases. Cells left blank indicate that that variable was not included. Changes in odds of MDR-TB (among TB cases) are shown for each 20% proportionate increase in the explanatory ...


Our study demonstrates that the high MDR-TB burden shown by aggregate statistics in Moldova masks substantial spatial heterogeneity18. Identifying areas of high burden and risk of MDR-TB should help prioritize resources and allow for targetted studies aimed at understanding the local drivers of emergence and spread of MDR-TB19. We find that the spatial variation in MDR-TB burden is striking; the estimated MDR-TB incidence rate in some rayons is 20 times that of others and the percentage of new and previously treated TB cases with MDR-TB ranges from 3% to 41% and 41% to 79%, respectively. Reports from regional-level surveys in some countries indicate that this degree of spatial heterogeneity may be characteristic8,20 and suggests that micro-epidemics could be occurring although more detailed molecular epidemiological studies are needed to confirm the underlying causes of these spatial patterns. Our study is the first to demonstrate this phenomenon across an entire country and to produce detailed maps showing that substantial heterogeneity may exist even within the boundaries of relatively small administrative regions (e.g. rayons).

That 23.5% of new cases had MDR-TB indicates that transmission of MDR-TB has a substantial role in this epidemic. Young, urban residents are at particularly high risk of resistance as are new TB cases with HIV co-infection. While the association between HIV and MDR-TB varies by setting21,22, this result concurs with univariable analyses from other FSU countries8,20,23. Our analysis is the first to adjust for multiple potential confounders providing robust evidence for this association, at least in this setting.

Two-thirds of all TB cases diagnosed in Moldovan prisons had MDR-TB and the estimated annual incidence was at least 570 (possibly as high as 1,300)/100,000 people. We believe this is the highest estimated incidence of MDR-TB ever reported. While it is unsurprising that MDR-TB rates are higher in prisons7,24, the percentages with MDR-TB are greater than those from other studies among prisoners in the FSU23,25. While improvements have been made within Moldovan prisons, case detection and use of TB diagnostics appear less accessible within the Transnistrian penitentiary system suggesting that there may be a substantial burden of undiagnosed drug resistance.

We found that a higher local risk of MDR-TB was associated with an increase in the percentage of TB cases that had previously been in detention; this suggests that the high rates of MDR-TB in prisons may be disproportionately contributing to increased local rates in the civilian sector. This could occur if prisoners become infected with MDR-TB whilst in detention, develop active disease following release, and subsequently spread these highly resistant strains in the community26. Our finding that cases who had previously been detained were at an increased risk of MDR-TB also supports this potential explanation. Previous studies have concluded that prisons may have a substantial role in fuelling TB epidemics at the population-level24,27 (i.e. across a geographic area and not just for an individual case) and our study provides additional data to support this theory.

A strength of this study is the high percentage of culture positive TB cases receiving quality-assured DST (96% by 2010). High coverage with diagnostics and a detailed patient database have allowed us to document the MDR-TB situation in Moldova with high resolution. Given similar historical approaches for TB control used in other FSU countries, the insights generated in Moldova may contribute to improved understanding of the resistance in this global region where MDR-TB is most concerning.

An important limitation of our study is our inability to fully assess the contribution that nosocomial transmission may have to MDR-TB incidence. Similar to many settings in the region, all tuberculosis patients in Moldova initiate treatment within hospitals. Additional molecular epidemiological studies within these settings20,28 are needed and will have implications for both infection control within hospitals and the use of ambulatory treatment.

Our study is also limited by the programmatic nature of the data and the reliance on existing approaches for TB case detection. It is possible that the quality of TB case detection varies by location and could partially explain the observed heterogeneity in MDR-TB risk. However we consider this unlikely to have had a substantial impact since Moldova is a small country with little geographic variation in terrain and well-distributed points to access TB care. Only about half of all notified pulmonary TB cases had positive cultures. Possible explanations for relatively low culture positivity rates include sub-optimal specimen collection, transport, or handling or incomplete entry of updated laboratory results into the database. Our estimates within the Transnistrian penitentiary system are especially prone to bias due to the relatively poor utilization of culture and DST.

While our locality-level model helped us to identify the presence of higher proportions of previously detained individuals as potential contributors to higher local MDR-TB risk, the model did not explain a large percentage of the local variability in resistance. This demonstrates that currently available data are not sufficient to explain local patterns in MDR-TB and further studies are needed to provide insight into why specific areas bear relatively high burdens and risk. Nonetheless, the spatial maps produced here provide information that can inform at least two types of action. First, identifying hotspots of burden and risk of MDR-TB permits efficient deployment of interventions that are often in limited supply to areas where they are most urgently needed. Secondly, these hotspots serve as ideal places to locate studies to identify local causes of resistance and to trial novel interventions.

MDR-TB presents a serious threat to public health in many global settings; to date the scale of response to highly drug-resistant disease has failed to meet the actual need4,29. This study underscores the role that comprehensive surveillance spatial analysis of MDR-TB and can play in understanding the local epidemiology of MDR-TB and improving responses to this crisis.

Supplementary Material



We thank all those involved in surveillance, laboratory testing and treatment for tuberculosis in the Republic of Moldova. This work was supported by Award Number U54GM088558 from the National Institute of General Medical Sciences. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of General Medical Sciences or the National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. MZ, AD and MD are staff members of the World Health Organization (WHO). The authors alone are responsible for the views expressed in this publication and they do not necessarily represent the decisions or policies of WHO.


1. World Health Organization. Global Tuberculosis Control 2011. Geneva: 2011. Report No.: WHO/HTM/TB/2011.16
2. Gandhi NR, Nunn P, Dheda K, et al. Multidrug-resistant and extensively drug-resistant tuberculosis: a threat to global control of tuberculosis. Lancet. 2010;375:1830–43. [PubMed]
3. Espinal MA, Kim SJ, Suarez PG, et al. Standard short-course chemotherapy for drug-resistant tuberculosis: treatment outcomes in 6 countries. JAMA. 2000;283:2537–45. [PubMed]
4. Zignol M, van Gemert W, Falzon D, et al. Surveillance of anti-tuberculosis drug resistance in the world: an updated analysis, 2007–2010. Bulletin of the World Health Organization. 2012;90:111–9. [PubMed]
5. Skrahina A, Hurevich H, Zalutskaya A, et al. Alarming levels of drug-resistant tuberculosis in Belarus: results of a survey in Minsk. Eur Respir J. 2011 doi: 10.1183/09031936.0145411. [PMC free article] [PubMed] [Cross Ref]
6. World Health Organization. Towards universal access to diagnosis and treatment of multidrug-resistant and extensively drug-resistant tuberculosis by 2015: WHO progress report 2011. Geneva: 2011. Report No.: WHO/HTM/TB/2011.3
7. Dara M, Kluge H. Roadmap to prevent and combat drug-resistant tuberculosis. Copenhagen: WHO Regional Office for Europe; 2011.
8. World Health Organization. Multidrug and extensively drug-resistant (M/XDR-TB). 2010 Global Report on Surveillance and Response. Geneva: 2010. Report No.: WHO/HTM/TB/2010.3
9. International Organization for Migration. Migration in Moldova: A Country Profile 2008. Geneva: 2008.
10. Crudu V. Republic of Moldova, 2006. Chisinau: 2009. Anti-tuberculosis drug resistance surveillance. Available from
11. Soltan V, Henry AK, Crudu V, Zatusevski I. Increasing tuberculosis case detection: lessons from the Republic of Moldova. Bull World Health Organ. 2008;86:71–6. [PubMed]
12. [Accessed 2 February 2012];2004 Population Census Results. 2004 at
13. World Health Organization. Guidelines for drug resistance surveillance in tuberculosis. 4. Geneva: 2009. Report No.: WHO/HTM/TB/2009.422
14. Shepard D. A two-dimensional interpolation function for irregularly-space data. 23rd Association for Computing Machinery National Conference; 1968; New York, USA. 1968. pp. 517–24.
15. Anselin L. Local Indicators of Spatial Association - LISA. Geographical Analysis. 1995;27:93–115.
16. Cox D. Some statistical methods connected with series events. Journal of the Royal Statistical Society Series B - Methodological. 1955:129–64.
17. Brooker S, Hay SI, Bundy DA. Tools from ecology: useful for evaluating infection risk models? Trends Parasitol. 2002;18:70–4. [PMC free article] [PubMed]
18. Cohen T, Colijn C, Wright A, Zignol M, Pym A, Murray M. Challenges in estimating the total burden of drug-resistant tuberculosis. Am J Respir Crit Care Med. 2008;177:1302–6. [PMC free article] [PubMed]
19. Cohen T, Manjourides J, Hedt-Gauthier B. Linking Surveillance with Action against Drug-resistant Tuberculosis. Am J Respir Crit Care Med. 2012 [PMC free article] [PubMed]
20. Wright A, Zignol M, Van Deun A, et al. Epidemiology of antituberculosis drug resistance 2002–07: an updated analysis of the Global Project on Anti-Tuberculosis Drug Resistance Surveillance. Lancet. 2009;373:1861–73. [PubMed]
21. Wells CD, Cegielski JP, Nelson LJ, et al. HIV infection and multidrug-resistant tuberculosis: the perfect storm. J Infect Dis. 2007;196:S86–107. [PubMed]
22. Suchindran S, Brouwer ES, Van Rie A. Is HIV infection a risk factor for multi-drug resistant tuberculosis? A systematic review. PLoS One. 2009;4:e5561. [PMC free article] [PubMed]
23. Dubrovina I, Miskinis K, Lyepshina S, et al. Drug-resistant tuberculosis and HIV in Ukraine: a threatening convergence of two epidemics? Int J Tuberc Lung Dis. 2008;12:756–62. [PubMed]
24. Stuckler D, Basu S, McKee M, King L. Mass incarceration can explain population increases in TB and multidrug-resistant TB in European and central Asian countries. Proc Natl Acad Sci U S A. 2008;105:13280–5. [PubMed]
25. Ruddy M, Balabanova Y, Graham C, et al. Rates of drug resistance and risk factor analysis in civilian and prison patients with tuberculosis in Samara Region, Russia. Thorax. 2005;60:130–5. [PMC free article] [PubMed]
26. Reyes H, Coninx R. Pitfalls of tuberculosis programmes in prisons. BMJ. 1997;315:1447–50. [PMC free article] [PubMed]
27. Basu S, Stuckler D, McKee M. Addressing institutional amplifiers in the dynamics and control of tuberculosis epidemics. Am J Trop Med Hyg. 2011;84:30–7. [PMC free article] [PubMed]
28. Nodieva A, Jansone I, Broka L, Pole I, Skenders G, Baumanis V. Recent nosocomial transmission and genotypes of multidrug-resistant Mycobacterium tuberculosis. Int J Tuberc Lung Dis. 2010;14:427–33. [PubMed]
29. Keshavjee S, Farmer PE. Picking up the pace--scale-up of MDR tuberculosis treatment programs. N Engl J Med. 2010;363:1781–4. [PubMed]
30. Agresti A, Coull BA. Approximate is better than “exact” for interval estimation of binomial proportions. The American Statistician. 1998;52:119–26.