Discussion
The current study uses information technology systems to examine the prevalence of diabetes (known and unrecognized) in all inpatients above age 54 years, in real time. In this cohort of 5082 patients, aged 54 years and above, 5% had unrecognized diabetes and 29% had pre-existing diabetes. Furthermore, almost 6% of all inpatients aged ≥54 years had poor glycemic control with HbA1c ≥8.5% (69 mmol/mol). After adjusting for age, gender, Charlson Index score, eGFR, and Hb levels, with admission unit treated as a random effect, patients with previously known diabetes had lower 6-month mortality. However, there were no significant differences in proportions of ICU admission, mechanical ventilation, or readmission within 6 months between the three groups.
The use of fasting plasma glucose for diabetes diagnosis is limited given the high prevalence of stress hyperglycemia and frequent use of glucocorticoid medications in inpatients18; oral glucose tolerance test for inpatient diabetes screening is resource intensive and impractical.19 HbA1c testing is superior for the inpatient diagnosis of diabetes1; however, sensitivity may be lowered by significant renal impairment, anemia, blood transfusions, and hemoglobinopathies, while prolonged stress hyperglycemia may produce false-positive results.20 ,21 These conditions are unlikely to be in numbers large enough to affect overall screening.22 HbA1c has the advantage of aiding both diabetes diagnosis and management.
Following adjustments for age, gender, Charlson Index score, eGFR, and Hb levels, with admission unit treated as a random effect, patients with previously known diabetes had lower 6-month mortality. There was no difference in mortality between patients with unrecognized diabetes and those with no diabetes. There is variation in the literature regarding diabetes being an independent predictor of mortality in inpatients. Vincent et al12 found no significant difference in mortality in the intensive care setting between patients with a history of insulin-requiring diabetes mellitus. Diabetes was associated with increased mortality in patients with community-acquired pneumonia in one study23; however, there was no association between diabetes and increased mortality in patients with community-acquired bacteremia,13 severe sepsis,12 trauma, burns, and acute cardiac failure.24 However, in the aforementioned studies, the presence of diabetes was determined from clinical history. In the current study, HbA1c as well as clinical records was used to identify patients with diabetes. In patients with congestive cardiac failure, those HbA1c <6.4% (46 mmol/mol) had higher all-cause mortality compared with those with HbA1c ≥8.7% (72 mmol/mol).25 Other studies have found a U-shaped association between HbA1c and mortality, with the lowest HR at HbA1c 7.5% (58 mmol/mol).26 The observational nature of this study means that we are unable to infer causality regarding the differences observed between groups. As yet we have no explanation for this surprising finding.
Electronic medical records and clinical information systems are increasingly utilized to identify and manage patients at risk. Examples include chronic kidney disease27 and colon cancer28 ,29 screening; ‘sepsis sniffers’ are an automated tool for the early identification and treatment of inpatients with acute sepsis.30 To our knowledge, the current study is the first reported example of large-scale programme to identify and treat patients with unrecognized and poorly controlled diabetes in an acute inpatient setting. In this large study, we investigated the clinical utility of routine HbA1c measurements in hospital inpatients. We demonstrate the use of electronic systems for aiding clinical care and early biochemical identification of a chronic disease; specifically using a combination of HbA1c results and the hospital medical records, we have demonstrated a method for the identification of patients with previously unrecognized diabetes as well as those with poorly controlled diabetes. This utilizes a previously missed opportunity of engaging with patients who are in hospital for non-diabetes-related conditions. Previous studies suggest that many patients with unrecognized diabetes do not have general practitioners.1 Interestingly, general practitioner rates were similar in the group of patients with unrecognized diabetes compared with those with no diabetes. This highlights the role of determining the glycemic status of inpatients for identifying patients who may otherwise remain unrecognized (and untreated). The utility of early diagnosis of diabetes in preventing diabetes-related complications depends on communicating this information to patients’ local doctors. One of the benefits of using electronic medical record to identify these patients is that electronic discharge summary can be used to communicate relevant information to patients’ local doctors automatically.
The prevalence of unrecognized diabetes has previously been reported to be 11% in a study of 2360 patients in another state of Australia in 2011.1 The lower proportion of unrecognized diabetes in our study may reflect increased diabetes awareness or age, ethnic, and socioeconomic differences between the study populations. Unlike those published previously,8 ,31 following adjustments for age, gender, Charlson Index score, eGFR, and Hb levels, with admission unit treated as a random effect, we did not observe an increase in length of stay associated with diabetes. Furthermore, we used median length of stay as an outcome variable which more accurately represents the non-parametric nature of this variable.
Diabetes and comorbidity data were obtained from medical record coding which may underestimate true prevalence. However, manual audits of the 20% of the sample found similar results. As mortality was determined by composite of in hospital mortality as well as patients’ deaths notified to the hospital, it is possible that some deaths were unrecorded. The short follow-up period means that we were unable to demonstrate any benefits of identification of unrecognized diabetes diagnosis on long-term complications. Given the 6-month study period, the effect of seasonal variation was not analyzed. This study investigated patients aged ≥54 years, as previous data suggested that the prevalence of unrecognized diabetes would be significantly greater in older patients32; however, it would be interesting to see if these results would be affected by the inclusion of younger patients. The observational nature of this study means that we are unable to infer causality regarding the differences observed between groups.