This is the first study to evaluate the osmole gap as a screening test for toxic alcohol exposure that conforms to the STARD criteria for reporting studies of diagnostic accuracy [31
]. In this sample, an osmole gap cut-off of 10 derived using the Smithline and Gardner equation with ethanol coefficients of 1 and 1.25 resulted in a relatively high sensitivity (> 0.85) but a low specificity (< 0.50) and high NPVs for the identification of patients for whom antidotal therapy for toxic alcohol poisoning would be indicated. These results therefore indicate that although this is not an ideal screening test, the osmole gap does provide additional diagnostic information. Specifically, a NPV of 0.95 for an osmole gap < 10 indicates a very high probability that if a toxic alcohol has been ingested, the serum level is below the threshold for antidotal therapy. This analysis does not suggest that the osmole gap should be used in isolation to provide the basis for discharging a patient without further clinical investigation or evaluation. However, it does indicate that the osmole gap does provide some additional diagnostic and prognostic information in terms of the probability that a patient will need antidotal therapy or hemodialysis. Using other clinical information and laboratory data in conjunction with the osmole gap will increase the accuracy of the diagnosis.
Using equation 2 to account for the supramolar contribution of ethanol resulted in an increase in the specificity of the osmole gap without a significant reduction in sensitivity. Although a screening test with a sensitivity of 1.0 is most desirable in this clinical situation, achieving this would result in a low specificity and a corresponding high false-positive rate which could result in the unnecessary initiation of treatment in these patients. Although antidotal treatment with either intravenous ethanol or fomepizole is relatively benign, the cost-effectiveness of any screening test must incorporate all costs and potential risks of misdiagnosis and inappropriate initiation of treatment which might include unnecessary hospitalization, treatment, or transfer by air or road ambulance.
Of the three exposed patients who may have been falsely diagnosed as unexposed using an osmole gap cut-off of 10, two had an elevated osmole gap upon presentation to the emergency department and would likely have been correctly diagnosed as exposed. The third patient had a compelling history of significant ethylene glycol ingestion, an osmole gap of 9.8, and an anion gap metabolic acidosis when they arrived at the emergency room and would also likely have been correctly diagnosed based on other available information.
A survey of studies of diagnostic accuracy published in four major journals concluded that the methodologies employed were generally inadequate to answer the questions posed [32
]. The results of that survey lead to the development of the Standards for Reporting Diagnostic Accuracy (STARD) statement which consists of 25 criteria that should be adhered to when evaluating the diagnostic accuracy of a test [31
]. This is the first formal evaluation of the diagnostic accuracy of the osmole gap that: i) includes subjects with all levels of exposure to ethylene glycol or methanol; ii) evaluates the osmole gap compared to the gold standard; and iii) conforms to essentially all other STARD criteria.
One of the necessary components of any screening test evaluation is a clear and valid definition of what constitutes a positive test [23
]. Although Aabakken et al. proposed a threshold of 20 using a different equation to calculate serum osmolarity, this was based on 177 consecutive patients admitted to an ED (mean age 65 years) without any exposure to ethylene glycol or methanol [15
]. Given that the normal range for the osmole gap may be higher in patients older than 60 years of age and that the osmole gap is generally used in conjunction with other diagnostic information, we felt that it is unlikely that this threshold is applicable to the diagnosis of toxic alcohol poisoning [34
]. In support of this, although applying an osmole gap cut-off of 20 in our sample resulted in higher specificity relative to an osmole gap of 10, it had a lower sensitivity that corresponded to six additional false-negative diagnoses [15
]. This reduction in sensitivity is concerning in this clinical scenario, given the potential serious and fatal ramifications of a false-negative diagnosis.
The diagnostic performance of the osmole gap is related to both the cut-off of the test and serum level being detected. Because we were unable to find any other empiric evaluation of the most appropriate osmole gap cut-off for the diagnosis of toxic alcohol poisoning, we elected to evaluate all possible cut-offs using ROC curves, and then specifically evaluate the clinically accepted osmole gap threshold of 10. Additionally, there is limited evidence supporting the current recommended thresholds of ethylene glycol and methanol concentrations for initiating antidotal therapy and hemodialysis that we used in this analysis. If these thresholds are too conservative as has been suggested, more liberal treatment thresholds would improve the diagnostic performance of the osmole gap [35
Because the inclusion criteria applied in this study required that both ethylene glycol and methanol be measured by gas chromatography, some patients may have had a toxic alcohol level above the threshold for antidotal therapy that was not measured. The diagnostic performance of the osmole gap may therefore be poorer if applied to all patients with any suspicion of exposure, including cases where exposure may have been potentially ruled out, albeit erroneously, before specific serum concentration measurements are performed.
This highlights the importance of interpreting the result of the osmole gap as it relates to other available information to estimate the likelihood of a toxic alcohol exposure. Information pertaining to the ingestion history, clinical signs and symptoms (e.g. visual disturbances) and other laboratory data including pH, HCO3-
, anion gap and urinalysis should all be considered concurrently [36
]. However, this study only evaluated the diagnostic accuracy of the osmole gap alone measured within 24 hours of the first recorded laboratory measurement. Consequently, these results are likely an underestimate of the overall clinical utility of the osmole gap when evaluated in conjunction with other diagnostic information as early as possible following exposure.
This study has three primary limitations. First, it is limited by the use of retrospective data and the requirement that all laboratory tests be performed on serum obtained simultaneously; this resulted in the exclusion of 103 subjects. However, to avoid making clinical assumptions using only laboratory data, we pre-specified that all required laboratory parameters had to be measured simultaneously. Because these measurements were not necessarily performed on the first blood sample obtained upon presentation to the emergency room, we restricted this analysis to only laboratory data collected within the first 24 hours of the first available blood sample. The rationale for this restriction was to mimic admission results as closely as possible within the constraints of a retrospective study. Realizing the critical importance of the simultaneous evaluation of all laboratory parameters immediately upon admission, and the time of the ingestion when interpreting the osmole gap, a prospective study is required to specifically address this methodological issue.
The second limitation of the study is the potential for referral or ascertainment bias. This sample only included patients who had serum ethylene glycol and methanol levels measured, which might suggest a high pre-test probability of exposure. However, ethylene glycol or methanol was only detectable in 34/131 (26%) patients. This incidence of detectable toxic alcohol concentrations suggests that the referral bias was not extreme and that the pre-test probability of exposure in this sample of patients was likely only moderate.
Finally, there is the potential for work-up bias. If an osmole gap above a specific cut-off is used as a prerequisite to measuring toxic alcohols using gas chromatography, this would tend to overstate the sensitivity of the test. We cannot exclude this possibility, especially in patients with toxic alcohol concentrations just above the threshold for antidotal therapy. Although it is possible that some of the 51 patients who had ethylene glycol and methanol concentrations requisitioned but not measured may have had elevated toxic alcohol concentrations despite an osmole gap below 10, we believe that workup bias was not significant in this dataset for several reasons. First, gas chromatography was performed in 25 patients despite an osmole gap below 10 calculated using equation 1, the most commonly applied equation in clinical practice. Second, we are not aware of any patient in either institution during the study period with an osmole gap less than 10 that subsequently developed significant toxicity. And finally, most of the cases with a detectable methanol concentration below the threshold for antidotal therapy had an osmole gap > 10.