| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Briefs |
(Department of Clinical Biochemistry, St. Boniface General Hospital, and University of Manitoba Medical School, Winnipeg, Manitoba, Canada;
aaddress correspondence to this author at: St. Boniface General Hospital, Winnipeg, Manitoba, Canada R2H 2A6; fax 204-231-2656, e-mail jkrahn{at}sbgh.mb.ca)
Abstract
Background: The osmolal gap (OG) is a screening test for the detection of toxic volatiles such as methanol and ethylene glycol. We used mean values of patient data to assess the diagnostic accuracy and long-term stability of OG measurements.
Methods: In a prospective study period in 2003, all requests for volatiles had OGs calculated and quality-control samples were analyzed for OG. ROC curves were constructed to determine whether OG could predict the presence of toxic volatiles in serum. This was also done in a retrospective study for data from 1996 to 2004. Our laboratory database was searched for all emergency room patients for the period of 1996 to 2004 who had tests ordered that allowed us to calculate OGs.
Results: For the prospective study period in 2003, the ROC areas indicated that we could accurately predict the presence of toxic volatiles but at markedly different decision cutpoints depending on the formula used. These cutpoints ranged from +10 to +33 mosmol/kg. In the retrospective study, the mean OGs in the patient population for each of the 3 formulas increased by 12 mosmol/kg from 1996 to 2004. For this reason, the diagnostic accuracy was poor when all data were analyzed together.
Conclusions: Under properly controlled conditions, the OG has high sensitivity and specificity for detection of poisoning with some volatiles. Over the long term, however, use of the reference interval of –10 to +10 mosmol/kg yields poor diagnostic accuracy because mean OGs are not constant over time. Bedside calculation is not advisable.
Calculation of serum osmolality from the serum concentrations of sodium, glucose, and urea is a long-standing practice (1)(2)(3). In earlier times, calculation of the osmolality gap (OG), defined as the difference between the measured and calculated osmolality, allowed clinicians to estimate the concentration of ethanol in a patient’s blood and provided evidence of suspected toxins, such as ethylene glycol and methanol, in the blood (4)(5)(6)(7). Because glucose and ethanol contribute more to osmolality than is inferred from their molar concentrations (8)(9)(10)(11)(12), the OG can be misleadingly high in the presence of increased glucose and will be higher than expected for the concentration of ethanol.
The performance of the OG in a diagnostic sense is unknown. There is no evidence that simple bedside calculations can be used to either rule in or rule out a poisoning. Formulas 1, 2 (12), and 3 (1) below accurately predict the measured osmolality (OSMm).
 | (1) |
 | (2) |
 | (3) |
We applied these formulas for calculated osmolality (OSMc) to data generated in our laboratory on control specimens and on patient samples from the emergency room, and calculated the OG for each OSMc.
Osmolality was measured by freezing-point depression with a Fiske 2400 osmometer (Fiske Associates). Electrolytes, glucose, urea, and ethanol were determined on high-volume analyzers [Hitachi 717 (1996–2002) and Roche Modular System (2003–2004); Roche Diagnostics]. Collection tubes were Becton Dickinson SST and PST Vacutainers from 1996 until 2002 and Corvac PST thereafter. The calculations were all done in SI units and expressed in mosmol/kg or mmol/L as appropriate. The respective OG values were calculated as OG = OSMm – OSMc. The prospective period of the study lasted from April to October of 2003. During this period, all requests for toxic volatiles also had all analytes measured to allow calculation of OG. We used 2 quality-control specimens (Bio-Rad MultiQual; Bio-Rad Laboratories) during that period and calculated OG on the control material.
We also carried out a separate retrospective search for all requests for toxic volatiles that also had the other laboratory tests performed to allow calculation of the OG. The practice in our region is that all requests for toxic volatiles are screened by an on-call clinical chemist. The data were carefully examined, and only diagnostic data were retained (without patient identifiers).
We performed ROC analysis on both the prospective and retrospective data and determined diagnostic specificity and sensitivity and areas under the curves (AUC). To determine whether the OG remained constant over long time intervals, we calculated mean OGs on data in 6-month intervals where sufficient tests had been ordered on emergency room patients to allow this calculation. This was done on data from 1996 to 2004. The statistical analysis was performed with the Microsoft Excel® add-in Analyze-It®. For the ROC analysis, any methanol or ethylene glycol >1 mmol/L was considered positive. To determine sources of variability in OG values, we assessed the data from quality-control specimens. The biggest contributions to the OSMc came from sodium and OSMm. It follows that imprecision in the OG results should be related to imprecision in sodium and OSMm. For both of these analytes as well as for OG, we calculated the difference from the mean value of that control specimen for each data point. This allowed us to do linear regression analysis of the differences in the OG vs the differences in OSMm and measured sodium.
During the prospective study, the quality-control samples that were run daily for the entire period demonstrated the stability of the analytical processes over that entire period. This study also revealed that the variability in the OG was largely attributable to the variability of OSMm, with little imprecision attributable to sodium (see the Data Supplement that accompanies the online version of this Technical Brief at http://www.clinchem.org/content/vol52/issue4). We do not have this type of data for the retrospective period.
Shown in Table 1
are the decision cutoffs at which 100% sensitivity was achieved, i.e., the OG value at which all toxic poisoning would be detected. ROC curves during the prospective study had high AUCs. Formulas 1–3 were able to correctly predict whether a toxic volatile was present; the decision cutoffs for formulas 1–3 were +10, +14, and +33 mosmol/kg, respectively, in agreement with their mean OGs of 0, +2, and +20 mosmol/kg in the study period. For the retrospective period, the AUCs varied between 0.79 and 0.80. The decision cutoffs to achieve 100% sensitivity were –9, –4, and + 7 mosmol/kg, respectively, for the 3 formulas.
|
The mean OG values for each 6-month period from 1996 to 2004 are shown in Fig. 1
. These statistics include all specimens (even those that contained ethanol). There is a difference of 20 mosmol/kg in the OG between the Dorwart formula and formulas 1 and 2. The Dorwart formula gave a mean OG of 8 mosmol/kg in 1996, which increased to 15–20 mosmol/kg in 2003–2004. Formulas 1 and 2 gave mean OG values of –12 to –13 in 1996, but had increased to 
0 to +2 mosmol/kg in 2003–2004. It is apparent that all formulas show a large shift in the mean OG over time. We have not been able to identify the reason for the change of the mean OGs. When we subjected these data to ROC analysis for each 6-month period, the decision cutoffs required to achieve 100% sensitivity showed the same trend as the mean OG values. This meant that the assumption of a reference interval of –10 to +10 mosmol/kg was wrong and therefore was not reliable. Throughout the period during which the mean OG values changed, the laboratory’s proficiency testing programs showed acceptable performance of the constituent analytes.
|
This study shows that OSMc and OG formulas can be a highly effective screening method, over a relatively short time period of a few years when careful quality-control procedures are followed, for the identification of patients poisoned with toxic volatiles. However, the drift in mean OG values over a longer time reduces the effectiveness and may account for the proliferation of formulas. This finding limits bedside calculations based on a reference interval of –10 to +10 mosmol/kg. It is evident that these calculations are effective only if they are validated on appropriate reference populations and if strict quality procedures are subsequently followed to ensure that the calculations continue to function in the intended manner.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
, OSMC1.86; 
, OSMC2; 
, OSMC-DORWART (formulas 1–3 in the text).