Article Text

Download PDFPDF

Increased risk of tuberculosis disease in people with diabetes mellitus: record-linkage study in a UK population
  1. F Young1,
  2. C J Wotton2,
  3. J A Critchley3,
  4. N C Unwin4,
  5. M J Goldacre2
  1. 1Institute of Health and Society, Newcastle University, Newcastle Upon Tyne, UK
  2. 2Unit of Health-Care Epidemiology, Department of Public Health, Oxford University, Oxford, UK
  3. 3Division of Population Health Sciences and Education, St George's, University of London, Cranmer Terrace, London, UK
  4. 4The Faculty of Medical Sciences, The University of The West Indies Cave Hill Campus, Bridgetown, Barbados
  1. Correspondence to Fiona Young, Institute of Health and Society, Newcastle University, Medical Sciences New Building, Richardson Road, Newcastle Upon Tyne NE2 4AX, UK; fiona.young{at}


Background The authors aimed to determine whether, and by how much, diabetes mellitus (DM) increases the risk of tuberculosis (TB) and conversely whether TB increases the risk of DM.

Methods Retrospective cohort analyses using data from two Oxford Record Linkage Study (ORLS) datasets, containing information on hospital admissions and day-case care between 1963 and 1998 (ORLS1) and between 1999 and 2005 (ORLS2), were carried out. The rate ratio (RR) for tuberculosis after admission to hospital with diabetes and for diabetes after hospital admission with tuberculosis was calculated.

Results In ORLS1, the RR for TB in people admitted to hospital with DM, comparing the latter with a reference cohort, was 1.83 (95% CI 1.26 to 2.60), and in ORLS2 the RR was 3.11 (1.17 to 7.03). RRs for pulmonary tuberculosis (PTB) and extrapulmonary tuberculosis (EPTB) within ORLS1 were similar at, respectively, 1.80 (1.16 to 2.67) and 1.98 (0.88 to 3.92). In ORLS 2 the RR for PTB was 2.63 (0.91 to 6.30). In ORLS1, there was no indication that TB was a risk factor for DM (RR 1.12, 0.76 to 1.60). The ORLS2 dataset was too small to analyse whether TB led to DM.

Discussion DM was associated with a two- to threefold increased risk of TB within this predominantly white, English population. The authors found no evidence that TB increases the risk of DM. Our findings suggest that the risks of PTB and EPTB were both raised among individuals with DM. As DM prevalence rises, this association will become increasingly important for TB control and treatment.

  • Diabetes mellitus
  • tuberculosis
  • Epidemiology
  • relative risk
  • cohort
  • diabetes DI
  • tuberculosis

Statistics from


An association between diabetes mellitus (DM) and tuberculosis (TB) has long been suspected. It was described in the works of Avicenna, in around ad 1000, and by Richard Morton, in his 1694 Phthisiologia: A Treatise on Consumption.1 However, the evidence base for this association is limited, and awareness of the association is poor. For example, current global and UK treatment guidelines for TB and DM make little mention of it,2–17 even though it is possible that it could have a large impact on tuberculosis control18 and treatment outcomes,19 especially as DM prevalence continues to rise globally.20

A meta-analysis of three prospective cohort studies, two of which were in patients with renal failure, described a rate ratio (RR) for pulmonary tuberculosis (PTB) among DM patients of 3.11.21 This increased risk of developing active disease is thought to occur as a result of an impairment in the immune response among individuals with DM.22–24 Diabetes is thought to be found more commonly in association with PTB than extrapulmonary forms of tuberculosis (EPTB),25 but evidence on this is limited. It has been postulated that the association between DM and TB is bidirectional and that TB can lead to an increased risk of DM.26 Few studies which address the direction of the association exist.27 There is some limited evidence that in active pulmonary disease, glucose tolerance is adversely affected,28 but it is not known whether, in the longer term, risk of DM is raised.

We aimed to establish whether individuals with diabetes within a UK population have an increased subsequent risk of developing tuberculosis disease (all forms), PTB or extrapulmonary tuberculosis and the magnitude of any such associations. We also aimed to establish whether the converse of this is true, if people with tuberculosis have an increased subsequent risk of developing diabetes and, if so, by how much.


Population and data

Data used for these analyses came from the Oxford Record Linkage Study (ORLS). The ORLS database is described in detail elsewhere.29 30 In brief, it contains statistical records of all NHS hospital admissions and all deaths occurring in defined populations in the former Oxford NHS region. This comprises two datasets, one for admissions between 1963 and 1998 (‘ORLS1’) and the second for admissions between 1999 and 2005 (‘ORLS2’, linked and built as the Oxford subset of English national Hospital Episode Statistics). The data items available for linkage changed between 1998 and 1999, and the two datasets are not themselves linked together. The data were assembled from routine NHS admission statistics (data similar to Hospital Episode Statistics from 1963 to 1987, and the actual Hospital Episode Statistics system from 1988), and exclude private-sector admissions (which are negligible). Episodes of day-case care as well as overnight stays were included in the ORLS from its outset. The hospital data are also linked to death certification data. Linkage is carried out using, notably, the national universal NHS number which was encrypted by the national supplier of hospital episode statistics and of death registration data using the same encryption method for both datasets.

The population covered by ORLS changed over time: it covered 350 000 people from 1963, 850 000 from 1965, 1.9 million from 1974 and 2.5 million from 1987 (all four counties, Oxfordshire, Buckinghamshire, Berkshire and Northamptonshire, covered by the former Oxford NHS Region).

Selection of cohorts

Five cohorts were constructed for the analyses in this study: one each for TB (all forms), pulmonary TB, extrapulmonary TB, DM (all forms) and a reference cohort (described in detail below). The basic methods were the same for each analysis and are described for the DM cohort and subsequent TB (all forms). The cohort was constructed by selecting the first record on file for each individual with DM, and the files were then searched for a subsequent record of TB. Records of people who had been admitted for various common orthopaedic, dental, ENT and other relatively minor disorders were selected as a reference cohort. This cohort has been constructed for use in other similar studies of disease associations.29 31 32 We followed the standard epidemiological practice, when hospital controls are used, of selecting a diverse range of conditions, rather than relying on a narrow range (in case the latter are themselves atypical in their risk of subsequent disease). As a check, we have studied the risk of DM and TB in each of the control conditions within the reference cohort, individually, to ensure that the reference cohort does not include control conditions that have atypically high or low DM or TB rates. This cohort was used to calculate expected numbers of people with TB in the DM cohort. A reference cohort was used, rather than using rates from an external population, to account for migration into or out of the Oxford region over the time period of interest. The use of a reference cohort, for this purpose, means it is not possible to calculate absolute disease occurrence rates, and instead, we calculated relative disease rates expressed as rate ratios comparing the DM cohort with the reference cohort.

Statistical methods

We took the date of entry into each cohort as the date of first admission for DM or comparison conditions, and the date of exit as the date of subsequent admission for TB, date of death or ending date for the dataset (1998 ORLS1, 2005 ORLS2), whichever was the earliest. We calculated rates of TB in the diabetes and reference cohorts combined, stratified by age (in 5-year groups), sex, district and single calendar year, and applied the stratum-specific rates for the combined cohort to the number of people in each stratum in, first, the DM cohort and, second, the reference cohort. This gave stratum-standardised rates for each cohort, and we calculated the ratio of the rate in the diabetes cohort relative to that in the reference cohort. The CI for the rate ratio and χ2 statistics for its significance were calculated using methods as described by Breslow and Day.33 We took p<0.05 as the level of statistical significance.

Further selection criteria

Selection into the DM cohort, and to the reference cohort, was confined to people with the conditions recorded as the main reason for hospital admission (to avoid selecting people who may have had other, more major diseases, and in whom DM or the reference condition had been recorded as incidental). We included people with TB regardless of position on the record. Anyone with both DM and a reference condition was assigned to the DM cohort. We excluded anyone with a previous record of TB. Anyone with TB on the same record as a first admission for DM was also excluded. We also did a subanalysis in which we excluded people with TB within a year of a first admission for DM. The latter two exclusions were done to minimise the possibility of Berksonian bias (a potential problem in hospital-based studies, as a result of the possibility that individuals with both conditions are more likely to be hospitalised than those with only one) or surveillance bias. We reversed all these procedures in studying DM after TB: TB was taken as the exposure cohort, diabetes was taken as the outcome, and stratum-specific rates of diabetes were calculated in the TB exposure cohort and compared with those in the reference cohort.

Where records had missing values for the stratification variables (age, sex and year of admission), they were excluded from the analysis. Data on each variable were missing from less than 1% of the records: age was missing for 0.4% of people in ORLS1 and 0.04% in ORLS2, sex was missing for 0.3% and 0.5% respectively, and year of admission was missing for 0.002% and 0.003% respectively.

All individuals in the ORLS database meeting the inclusion criteria for this study were included in the analyses, so the overall study size was fixed by the size of the ORLS population, the number meeting the inclusion criteria and the number of years for which data were available.


In the ORLS1 dataset, there were 19 244 patients in the DM cohort, 6997 in the TB cohort and 572 131 in the reference cohort. The mean age of entry to the diabetes cohort was 52 years, with a mean follow-up of 7.1 years. In the TB cohort, the mean age at entry was 49 years, and the mean follow-up was 11.2 years. In the reference cohort, the mean age at entry was 31 years, with a mean period of follow-up of 11.6 years.

In the ORLS2 dataset, there were 7943 patients in the DM cohort, 999 in the TB cohort and 230 085 in the reference cohort. The mean age of entry into the diabetes cohort was 51 years, with a mean follow-up of 3.2 years.

In the TB cohort, the mean age at entry was 46 years, and the mean follow-up was 2.4 years. In the reference cohort, the mean age at entry was 39 years, with a mean period of follow-up of 3.5 years.

Matching ratios—the number of people in the reference cohort per person in the DM cohort—varied by age from 7:1 to 182:1 in ORLS1 and from 15:1 to 109:1 in ORLS2. For TB, the corresponding matching ratios varied by age from 31:1 to 526:1 in ORLS1 and from 104:1 to 1577:1 in ORLS2. Age distributions for the ORLS1 and ORLS2 cohorts are given in table 1.

Table 1

Age distributions for the Oxford Record Linkage Study 1 and Oxford Record Linkage Study 2 study populations of people admitted to hospital with diabetes mellitus and tuberculosis: number and percentage of people in each age group at time of admission

We found significantly raised rate ratios for TB (all forms) among individuals with diabetes both in the 1963–1998 ORLS1 dataset and in the 1999–2005 ORLS2 dataset. The rate ratios were, respectively, 1.83 (95% CI 1.26 to 2.60) and 3.11 (1.17 to 7.03) (table 2).

Table 2

Occurrence of tuberculosis in people admitted with diabetes mellitus, and the converse: number of people in the reference cohort* with either outcome, observed and expected number of people with outcome condition in the exposure cohort, ratio of rate in the exposure cohort to that in the reference cohort, 95% CIs and p values for the rate ratio

There was no significant elevation of risk of DM after TB in the ORLS1 dataset, 1963–1998, with a rate ratio of 1.12 (0.76 to 1.60) (table 2). The numbers in ORLS2 were too small for a meaningful analysis of the risk of DM after TB.

Although our principal analyses were based on patients whose record specified DM as the main diagnostic reason for admission, we also analysed all combinations of diabetes and TB as main or secondary reasons for admission. This did not affect the results materially. Including all records of diabetes, in secondary as well as main diagnostic positions, the rate ratio for tuberculosis after diabetes was 1.77 (1.45 to 2.15, based on 131 observed cases) in ORLS1 and 2.56 (1.78 to 3.69, based on 59 observed cases) in ORLS2.

Rate ratios for diabetes after TB were not significantly raised. We found an increase in the rate ratio for PTB among individuals with diabetes compared with those without diabetes both in the 1963–1998 ORLS1 dataset and in the 1999–2005 ORLS2 dataset at, respectively, 1.80 (1.16 to 2.67) and 2.63 (0.91 to 6.30) (table 2). In ORLS1, there was a non-significant elevation of risk for extrapulmonary tuberculosis after DM, with a rate ratio of 1.98 (0.88 to 3.92). In ORLS2, the numbers were too small for meaningful calculation of rate ratios (table 2). All figures given have been adjusted for Berksonian bias (all diagnoses within a year of first admission were removed from analyses).

Because the incidence of tuberculosis has changed very substantially in the period covered by the study, as background contextual information we analysed temporal trends in person-based admission rates for tuberculosis in the area. They were 56 per 100 000 population in 1964, 26 in 1974, 13 in 1984, four in 1994 and five in 2004.


As far as we know, this is the first study of the association between DM and TB in a UK population to address both the magnitude and direction of the association.21 27 We found an increased risk of developing active tuberculosis in people with DM. Our analyses have shown that the association is unlikely to be bidirectional with no evidence of increased risk of developing DM among individuals with TB. The risk of PTB was significantly raised among individuals with DM, but our findings for the risk of developing EPTB among individuals with diabetes were not significant. However, numbers in the latter analysis were small, with low statistical power. Although we found no indication that TB leads to DM, it should be noted that it is known that some TB treatments can cause transitory hyperglycaemia.34

The methods used in this study have both strengths and weaknesses, the long period of follow-up in ORLS1 and the size of both datasets being strengths. A limitation in using the ORLS datasets is the inability to control for some of the possible confounding or explanatory factors for the association between DM and TB. Although age, sex and district of residence are controlled for, alcohol intake, smoking, ethnicity and socio-economic status are not. However, the population covered by the ORLS lives mainly in market towns, rural areas or relatively small conurbations such as Oxford, Reading and Northampton. It is generally affluent and healthy, compared with the English national average. In principle, ethnicity could be a confounder, as some populations are known to have an increased risk of developing both TB and diabetes independently. However, of the people in the ORLS1 dataset, 55% were born in the Oxford region itself and 94% in the British Isles (equivalent data not available for ORLS2), and thus the scope for confounding by ethnicity is likely to be small. This study uses hospital admission data (including day cases): accordingly, there is a potential for selection bias and incomplete follow-up. The majority of individuals with TB are likely to have a hospital episode and be captured in the dataset. However, those with diabetes may never be admitted to hospital, and those who are admitted may have more severe diabetes and poor glucose control. Administrative hospital data may also underestimate the occurrence of diabetes, particularly if diabetes is present but not the main reason for hospital admission. There is limited evidence, from an elderly cohort, that poor blood glucose control is related to an increased risk of TB.21

A systematic review found only 13 studies that examined the association between diabetes and tuberculosis, most of which were not specifically designed to do so; these studies are explicitly compared within this review.21 In brief, the majority of the studies were cross-sectional or case control in design with only a single prospective population-based study reported; TB and DM diagnosis was often based on case reports.21 27 Thus, our study adds to a relatively small evidence base. Our finding that diabetes is associated with an increased risk of TB is consistent with the findings in these published studies. Our analyses further the available evidence by addressing the direction of the association between TB and diabetes, and investigating separately the risk of PTB and EPTB in individuals with diabetes. Our study is also novel in that it addresses the association in a mainly ‘white British’ population. The largest study to have addressed this association was carried out in Korea by Kim et al where an overall RR of 3.5 (95% CI 3.0 to 4.0) for PTB among individuals with diabetes was found. The only study to address this association in a UK population, a case control study of TB cases identified between 1990 and 2001 carried out by Jick et al, found an OR of 3.8 (95% CI 2.3 to 6.1) for TB in individuals with diabetes. These findings are comparable with the rate ratio we found for PTB in individuals with diabetes in ORLS2, although somewhat higher than the RR calculated within ORLS1.

We cannot account definitively for the difference in magnitude of risk of TB comparing ORLS1 and ORLS2. It is noteworthy, however, that TB was much more common in the early period of ORLS1 than in the later period of ORLS2. We speculate that, at a time when TB was common, the risk factor of diabetes was less apparent, ‘diluted’ by other environmental factors, and that, at a time when TB became uncommon, the risk factor of diabetes became more evident.

The implications of these findings within a UK context are likely to become more important as the prevalence of diabetes rises. Attempts at effective TB control may be hampered. Policy should guide local health departments so that they can identify patients with TB promptly and start them on the appropriate treatment without delay. This would reduce the transmission of tuberculous disease and the number of contacts per index case. People with DM could become important subjects for screening for TB, dependent on the outcome of cost–benefit analysis of screening, or for interventions such as active case finding and treatment of latent TB. In areas such as India or Africa, where TB is still endemic, and the prevalence of diabetes is increasing, the potential implications for public health and clinical practice are substantial. It has been estimated that diabetes could account for approximately 15% of incident PTB in India, and around 20% of sputum-positive cases.18 There is evidence that individuals with concomitant TB and diabetes have poorer TB outcomes19 35 36; this needs to be addressed, as it is evident that this population of patients are at present poorly identified within treatment guidelines and are likely to increase in number as diabetes prevalence rises.

To further the investigation of the association between diabetes and TB, analyses could be carried out in datasets that contain more information on possible confounders, including ethnicity and socio-economic status, and behaviours such as smoking. However, to validate our findings beyond this, a large case control or cohort study would be needed. Assuming the association is genuine, further research on TB and DM is needed to describe the effectiveness of potential case finding programmes in people with diabetes and to develop the most appropriate treatment strategies for those with concomitant disease.

What is already known on this subject

  • Only 13 studies examine the association between diabetes and tuberculosis, most of which were not specifically designed to do so.

  • A meta-analysis of three prospective cohort studies (two of which were in patients with renal failure) calculates a RR for pulmonary tuberculosis (PTB) among DM patients of 3.11.

  • To our knowledge, no study has addressed whether this association is bidirectional.

What this study adds

  • We used the Oxford Record Linkage Study dataset to determine if individuals with diabetes have an increased risk of developing tuberculosis and the converse.

  • Our study adds to a relatively small evidence base finding a significant increase in the incidence of tuberculosis among individuals with diabetes within a UK population.

  • We also addressed the directionality of this association and found no association between TB and subsequent risk of diabetes.



  • Funding The Unit of Health-Care Epidemiology is funded to undertake record linkage studies by the NIHR Co-ordinating Centre for Research Capacity Development.

  • Competing interests None.

  • Ethical approval Ethics approval was provided by the Central And South Bristol Multi-Centre Research Ethics Committee (04/Q2006/176).

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.