The burden of influenza is difficult to estimate due to lack of specific outcome measures in administrative data. Using statistical regression techniques, we have distinguished the effects of seasonal influenza in working-age Manitoba adults with and without diabetes. Influenza was associated with increased rates of physician visits and hospitalisations for influenza-like illness in non-diabetic adults, accounting for 15% of ILI during influenza season. In diabetic adults, influenza was associated with increased ILI, PI and ALL, accounting for 13%, 26% and 6% of these outcomes, respectively. Compared with working-age non-diabetic adults, working-age adults with diabetes experienced a significantly greater 6% increase in influenza-attributable ALLs. Since a similar difference was not observed for ILI, we infer that adults with diabetes experience a disproportionately greater susceptibility to more serious manifestations of influenza, requiring hospitalisation.
In absolute terms, working-age adults with diabetes experienced an additional six hospitalisations due to influenza per 1,000 adults. Notably, if these same adults had not had diabetes they would not have experienced any ALLs due to influenza, since influenza did not appear appreciably to increase ALLs in those without diabetes.
Limitations and strengths
Our study has several limitations. First, because we relied on a community-level indicator for influenza, ecological bias may arise if outcomes attributed to influenza did not actually occur in patients infected with influenza. For example, respiratory syncytial virus (RSV) co-circulates with influenza and causes similar illness . Unaccounted RSV would lead to overestimates of influenza-attributable illness. Because the proportion of unaccounted RSV is not likely to be different in those with diabetes, our estimates of the relative effects of diabetes on influenza-attributable illness may not be affected. Indeed, we obtained similar results using a more specific measure of influenza activity—the proportion of respiratory specimens positive for influenza virus (data not shown).
Second, as the present study was observational, other unrecognised confounders may affect our findings. For example, obesity has emerged as a risk factor for hospitalisation and severe outcomes in pH1N1 cases [34–36]. It remains unclear from the pH1N1 literature whether obesity is an independent risk factor for influenza or whether the association between obesity and severe outcomes of influenza is itself confounded by diabetes, cardiovascular disease and other obesity-related comorbidities . One recent study of the NHANES cohort found a consistent association between obesity and pH1N1 hospitalisations/death, after stratifying by comorbidity, suggesting a true diathesis . In the seasonal influenza literature, Kwong et al have demonstrated an association between obesity and increased respiratory hospitalisations, after adjusting for diabetes and cardiovascular disease . Obesity may therefore account for the association between diabetes and influenza-attributable illness to an unknown extent . Other clinical variables we were unable to include were the specific type of diabetes; and variables related to diabetes severity and control, such as HbA1c. The incorporation of such variables in future studies would provide additional evidence helping to confirm or disprove a diabetes-specific effect. Our results likely generalise to those with type 2 diabetes, which comprises the vast majority of diabetes in adults. We adjusted for comorbidity using the number of ADGs instead of specific individual-level diagnoses, due to practical modelling limitations.
Finally, our data is limited to healthcare use, which may under-represent the total burden of influenza. Many working-age adults with influenza do not present for medical attention although they may still incur absenteeism and productivity losses [1, 39]. Previous studies have suggested that patients with diabetes who experience infectious diseases may be more likely to present for medical attention and to be hospitalised [40, 41], and this may lead to overestimates of influenza-attributable illness specifically in diabetic adults. Because increased medical attention likely applies generally, regardless of infectious aetiology (e.g. ILI, PI or ALL of both influenza and non-influenza aetiologies), we have accounted for this source of potential confounding by singling out differences in the effects of circulating influenza among diabetic and non-diabetic adults, while adjusting analytically for the effects of diabetes alone.
Three previous studies have compared influenza-attributable outcomes in patients with and without diabetes during periods of seasonal influenza. Schanzer et al noted increased influenza-attributable primary respiratory admissions in patients with diabetes compared with those without diabetes, but did not report diabetes-specific risk estimates . Bouter et al found that diabetes increased the risk of both pneumonia hospitalisations and subsequent mortality to a greater extent during years with significant influenza activity . Finally, Neuzil et al reported a fivefold (unadjusted) higher rate of influenza-attributable cardiopulmonary hospitalisations or deaths in working-age women with diabetes . The limitations of these studies include potential bias from the use of hospital-based comparison groups , lack of adjustment for comorbidities and vaccination status [11–13] and inadequate adjustment for seasonality , as mentioned earlier. Additionally, several studies have examined the effects of diabetes on pH1N1 influenza. Diabetes appears disproportionately represented in most studies of confirmed pH1N1 cases requiring hospitalisation [36, 42–45]. Some pH1N1 studies have also identified diabetes as a risk factor for severe outcomes following hospitalisation , although other studies have not [46, 47]. Whether the high prevalence of diabetes in pH1N1 studies can be attributed to a diabetes-effect on risk , confounding due to other comorbidities [46, 47] or due to adults with diabetes being more readily admitted on a precautionary basis  remains unclear. Moreover, since over one-third of diabetes is undiagnosed in the community , the lack of a population control group in case series of pH1N1 hospitalisations makes it difficult to ascertain whether the apparent effect of diabetes on pH1N1 hospitalisations is due to increased detection of diabetes in tertiary care. The prevalence of diabetes in pH1N1 patients may be consistent with its community prevalence if undiagnosed diabetes is included .
To our knowledge, our study is one of only two studies  to have identified and followed individuals, in the general population, with and without diabetes, for influenza-attributable outcomes. The use of administrative databases ensures comparable data quality among all study subjects. Our study used a validated case definition  to define diabetes status before outcome ascertainment, thus avoiding differential detection of undiagnosed diabetes in adults hospitalised for a study outcome. Moreover, ours is the only study to have adjusted for comorbidities, vaccination status and cyclic trends apart from influenza. Insofar as diabetes was found to be a risk factor for influenza-attributable hospitalisations, our results agree with those of previous seasonal influenza studies [11–13] as well as those of certain pH1N1 studies . In contrast to previous studies, the effect we demonstrated was much smaller and subtler in magnitude. These findings highlight the importance of appropriate adjustment for potential confounders and cyclic trends, which may otherwise lead the burden of seasonal influenza to be overestimated in high-risk populations. Our findings also suggest that we should be cautious of pH1N1 series showing a high prevalence of diabetes, although our data did not include 2009 and are therefore not straightforwardly applicable.
Clinical and policy implications
Our results suggest that working adults with diabetes experience greater risk of influenza-related ALLs. The public health impact of diabetes on the burden of influenza in Manitoba may be summarised as an additional six hospitalisations per 1,000 diabetic person-years. A previous economic analysis of a universal vaccination programme in Ontario estimated the cost of a pneumonia- or influenza-related hospitalisation to be $6,418 Canadian dollars (CAD), and the cost of a universal vaccination programme to be $7.55 CAD per dose delivered . Using these figures as rough approximations, it may be cost-saving to target working-age adults with diabetes for vaccination, with a presumed rate of vaccine effectiveness as low as 20%. Of course, the unit cost of a diabetes-specific vaccination programme may be much higher, and the actual effectiveness of influenza vaccinations in high-risk populations is not well established. While observational studies suggest that vaccination against influenza reduces hospitalisations in working-age adults with diabetes [10, 50], these benefits may be attributable to healthy user bias . Our observation that working-age adults with diabetes experience a greater burden of influenza than similar non-diabetic adults provides a clinical justification for targeted anti-influenza interventions; identifying particular interventions and evaluating their effectiveness in this population are questions for further research.
Vaccination guidelines indirectly single out working-age adults with diabetes for routine vaccination. We have demonstrated an increased burden of influenza in this population. Randomised trials are needed to confirm actual vaccine effectiveness in this group. Formal economic studies are also required, to ascertain the extent to which identifying diabetes as a high-risk indication for vaccination may mitigate the healthcare use and costs associated with influenza. Until such studies are available, our work represents the strongest current evidence highlighting the burden of influenza, and the potential benefits of influenza vaccination, in diabetic adults.