Data sources

We identified all patients 16 years old and above who were logged in PAS during the calendar year 2010 as hospital admissions, whether elective or non-elective. PAS data were linked to PICS data and patients with a recorded diagnosis of diabetes in the PAS or who were identified in PICS as having received treatment with anti-diabetic medication were excluded. This broad exclusion criterion was used because our main purpose was to determine the frequency of hypoglycemia that could not be explained by the use of prescribed hypoglycemic agents. We included patients admitted to the intensive care unit (ICU) in the denominator, as they invariably have a period of stay outside ICU (susceptible population). This identified a population of non-diabetic in-patients who could suffer a hypoglycemic episode outside the critical care setting.

We identified episodes of hypoglycemia in three ways. We directly identified low blood glucose concentrations from the PICS database, which captured both bedside and laboratory blood glucose estimations; we indirectly identified hypoglycemia from prescribed treatments for hypoglycemia whose administration was recorded in the PICS database; and we identified diagnostic codes for hypoglycemia from the PAS database. We regarded any one of these as a trigger event that signalled potential hypoglycemia. If a trigger occurred during a period of time the patient spent in ICU the patient's case was excluded from the numerator.

Cut-off value for hypoglycemia

Various blood glucose concentrations have been used to define hypoglycemia in non-diabetic patients. Previous studies have used 2.7 [4], [30], 3.0 [31] and 3.3 mmol/l [1], [11]. Meanwhile 2.2 mmol/l is used to define severe hypoglycemia in diabetic patients [26] and 2.5 mmol/l has been used for forensic investigations [32]. Considering the uncertainty, we analysed the data at different values of blood glucose concentration, to establish the effect on perceived occurrence. We did not differentiate between laboratory blood glucose values and point-of-care blood glucose values, or consider the type of equipment used to measure glucose values.

Medication as an indirect indicator of hypoglycemia

We examined electronic prescription records for medication used to treat severe hypoglycemia, to establish whether these may serve as triggers in detecting hypoglycemia in non-diabetic patients. The triggers extracted from PICS were intramuscular glucagon injection, intravenous glucose 10%, 20%, and 50% solutions; and oral glucose 40% gel. The case-notes of patients who received any of these but had neither a prescription for anti-diabetic medication in PICS nor a diagnostic code for diabetes in PAS were reviewed to establish whether they had in fact been hypoglycemic. Reasons for false positives were documented.

Discharge diagnostic codes for hypoglycemia as an indirect indicator

We examined ICD10 discharge diagnostic codes for hypoglycemia (E15, E16.0, E16.1, and E16.2) in the PAS system for the year 2010. The case-notes of patients with these ICD 10 codes who had neither a prescription for anti-diabetic medication in PICS nor a diagnostic code for diabetes in PAS were reviewed to determine the validity of the triggers in identifying non-diabetic hypoglycemia. Again reasons for false positives were recorded.

Electronic point-of-care and laboratory blood glucose concentrations as direct indicators of hypoglycemia

From available electronic observations derived from PICS, point-of-care or laboratory blood glucose values were used to identify patients who had been hypoglycemic during their in-patient spell. We reviewed the case-notes of patients categorised as non-diabetic by the criteria described above.

Determining the causes of hypoglycemia in non-diabetic patients

Information on diagnosis that was noted in the discharge diagnostic codes from PAS was first verified as an accurate description by case-note review. Any missed diagnostic codes were documented. A checklist of potential causes derived from previous literature was used to identify possible reasons for the hypoglycemia [1]–[13]. Because ICD 10 codes are limited in their ability to describe aetiology, for every patient with a blood glucose value less than or equal to 2.7 mmol/l, two investigators (KN and PN) made a judgment, based on the co-morbidity and patient condition, as to whether the patient may have had an unexplained hypoglycemia.

Statistical analysis

Data were analysed using STATA® 10 software [33]. Cases identified by any of the three triggers were used to estimate the incidence of hypoglycemia in non-diabetic patients. As none of the three data sources is complete we used the capture-recapture technique for three sources [34]. Capture- recapture methods have been used in health care to estimate population prevalence using multiple incomplete sources [35]–[40]. In summary, eight log-linear models, each specifying different interactions between the three sources, are derived to estimate the size of the total population. The layout of the three-source models is given in supporting information (Text S1). Calculations are based on the overlap between the three sources. The best estimate of the eight given is chosen using the Akaike information criteria (AIC) and the Bayesian information criteria (BIC). The 95% confidence intervals around the estimates were calculated using the goodness-of-fit based method [41]. An example of the working using the three source model is given in supporting information (Text S2) [42]. We estimated the frequency at different cut-off values: 3.3, 3.0, 2.7, 2.5 and 2.2 mmol/l. Frequency is reported as a count for the observed number of admissions of patients who did not have diabetes or receive diabetic medication in a non critical care setting for the year 2010. We repeated the same analysis stratifying the population by age into those 65 years and above and those who were less than 65 years old.

The study was considered by the Clinical Governance Support Unit of the University Hospital Birmingham (which has the function of an institutional review board to determine whether a study requires ethics approval and patient consent or is considered to be audit of clinical practice). The Unit approved the study as an audit that used routinely collected data for service improvement, involved no experiment, and did not require written patient consent.

The frequency of hypoglycemia among non-diabetic patients in non critical care setting

In combination the triggers identified 71 unique admissions with hypoglycemic episodes at a cut-off of 3.3 mmol/l, 59 at 3 mmol/l, 37 at 2.7 mmol/l, 30 at 2.5 mmol/l and 23 at 2.2 mmol/l. Each of these admissions was of a unique patient. Using the capture-recapture method at 3.3 mmol/l cut-off, an estimate of 189 (95%CI 124 to 352) hypoglycemic episodes is predicted in a non diabetic population of 37 898, giving a cumulative incidence of 50 per 10 000 admissions (95%CI 33–93). Estimated cumulative incidence at 3.0 mmol/l was 36 (95%CI 24–64), at 2.7 mmol/l, 13 (95%CI 11–19), at 2.5 mmol/l, 11 (95%CI 9–15) and at 2.2 mmol/l, 8 (95%CI 7–11) per 10 000 admissions (Figure 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Number of hypoglycaemia episodes –v- threshold blood glucose concentration (mmol/l) and upper and lower 95% confidence bounds per 10,000 admissions.

https://doi.org/10.1371/journal.pone.0040384.g002

Analysis showed admissions with an age above 65 years were more (approximately 50% more) likely to have an episode of hypoglycemia compared to the younger age group at all cut-off points. Estimated frequency above the age of 65 years at 3.3 mmol/l was 55 (95%CI 32–149), at 3.0 mmol/l, 39 (95%CI 24–158), at 2.7 mmol/l, 18 (95%CI 15–27), at 2.5 mmol/l, 15 (95%CI 13–23) and at 2.2 mmol/l, 13 (95%CI 11–21) per 10 000 admissions (Table S1).

Causes of non-diabetic hypoglycemia

Characteristics of the non-diabetic patients who had hypoglycemia at a cut-off point of 3.3 and 2.7 mmol/l are given in Table 1. Most patients (>90%) were admitted as an emergency. The commonest co-morbidities linked to hypoglycemia were sepsis, kidney disease, and alcohol dependence. Others included pneumonia, liver disease, cancer, and self-harm with hypoglycemic agents. Most patients had multiple possible reasons for their hypoglycemia (Table S2). Detailed case-note review of those with blood glucose concentrations less than 2.7 mmol/l revealed seven patients who did not have a plausible reason to explain the occurrence of hypoglycemia. However all seven were either admitted for investigation of hypoglycemia that occurred elsewhere or had an episode that was noted on admission; there was no unexplained hypoglycemia that occurred after admission during inpatient stay (Figure S1). Over 30% of patients whose blood glucose concentration was lower than 3.3 mmol/l, and nearly 40% of those whose blood glucose concentration fell below 2.7 mmol/l, died.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of the patients identified as non diabetic hypoglycemic patients.

https://doi.org/10.1371/journal.pone.0040384.t001

Summary of findings

Non-diabetic hypoglycemia is rare in hospital in-patients. We estimate that, with a cut-off value of 2.7 mmol/l, 13 (95% CI 11–19) episodes per 10 000 admissions occurred in one year; and with a cut-off value of 3.3 mmol/l 50 (95% CI 33–93) episodes per 10 000 admissions per year. Estimates are higher in patients above the age of 65 years (39 and 55 per 10 000 admission respectively at cut-off values of 2.7 and 3.3 mmol/l). All the cases of hypoglycemia that occurred after admission could be explained by co-morbid conditions, principally alcohol dependence, renal failure, and sepsis.

Comparison with other studies

Our estimates are similar to previous studies [4], [30], [31] except that of Mannucci et al [11] who reported an incidence of 8.6% in a small (678 patients) elderly population from a single medical unit. However the study was mainly based on routine blood glucose concentrations measured in the fasting state, defined hypoglycemia at a cut-off point of 3.3 mmol/l, and examined patients with a mean age of 81 years admitted with co-morbidities commonly associated with hypoglycemia.

Our study supports the previous observations that non-diabetic patients who develop hypoglycemia are more likely to die than those who do not [1], [11], [30]. Similarly, hypoglycemia in our study was often associated with renal disease, sepsis or pneumonia, and alcohol dependence, and other co-morbid diseases, in accord with previously suggested associations [8], [11], [13], [30]. Hypoglycemia can be a marker of poor prognosis, so that identified patients may need intense monitoring and management. If a cause for severe spontaneous hypoglycemia is not obvious, then the correct samples should be taken prior to treatment so that a diagnosis can be made. The samples can subsequently be analyzed for insulin, C-peptide, and other markers along with confirmatory laboratory glucose measurement and routine blood samples for renal disease, liver disease, and sepsis.

Limitations

Our study was retrospective. All three sources of information were incomplete, as we expected. We were able to overcome this by using the capture-recapture technique, and provide estimates of the true means with relatively narrow confidence intervals. Our study involves only one large hospital in UK. The hospital has specialist renal and liver units and electronic recording of near-patient blood glucose tests, which will have increased the observed frequency of hypoglycaemia. However, we accept that observed rates may be different elsewhere, and depend at least in part on the patient population and on the nature and frequency of testing.

Conclusion

Significant non-diabetic hypoglycemia in hospital in–patients (at or below 2.7 mmol/l) outside critical care is rare. It is sufficiently rare for occurrences to merit case-note review and diagnostic blood tests, unless an obvious explanation is found.