Methods

Using the search function of the LIS, we retrospectively identified all patients that had a serum potassium concentration ⩽2.5 mmol/l over three 6-month periods: prior to the introduction of reflective testing (“baseline”); after the introduction of reflective testing by a chemical pathologist but before reflex testing; and after the introduction of reflex testing. Using the first specimen with a potassium ⩽2.5 mmol/l as a reference point, patients who had magnesium assayed in the previous 6 months were excluded to avoid patients with known metabolic problems that involve magnesium. Also excluded were those <18 years of age, dialysis patients where hypokalaemia was almost certainly due to dialysis, and specimens thought to be contaminated with intravenous fluid. Serum albumin was not routinely measured on all samples, although this would be one consideration when deciding on treatment of hypomagnesaemia. All assays were performed by an Olympus AU2700 analyser (Olympus Life & Material Science Europa, Co. Clare, Ireland) using Olympus reagents according to the manufacturer’s instructions.

The proportion who had serum magnesium measured during that episode of hypokalaemia was determined and compared by means of a 2×2 χ² test using Microsoft Excel (see table 1). It was noted from where the request originated, that is, from the clinician or within the laboratory. The proportion who had hypomagnesaemia with serum magnesium concentration <0.78 mmol/l, and the proportion who had severe hypomagnesaemia with serum magnesium concentration <0.50 mmol/l, was also determined and compared using a 2×2 χ² test.

Results

Table 1 summarises sample characteristics and serum magnesium measurement patterns across all three periods. The number of severely hypokalaemic specimens increased over time, which is likely to reflect the increasing number of specimens received by the laboratory. Chronologically, there was a statistically highly significant increase in measurement of serum magnesium concentration on the initial specimen, initial or first subsequent specimen, or any specimen during that episode of hypokalaemia, following introduction of reflective and reflex testing when compared to the previous system (at least p<0.001 in all cases) (fig 1). The increase associated with introduction of reflective testing applied to requests from both clinicians and laboratory staff (fig 2), whereas after introduction of reflex testing, all requests were initiated automatically by the LIS.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Percentage of specimens with magnesium check. *p<0.001 compared to previous system.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Source of magnesium check. *p<0.001 compared to previous system.

Mean serum magnesium concentration for specimens where this was measured was 0.64 mmol/l (95% CI 0.49 to 0.78 mmol/l), 0.68 mmol/l (0.62 to 0.73 mmol/l) and 0.70 mmol/l (0.66 to 0.74 mmol/l) for each 6-month period chronologically. These differences were not significant when using a two-tailed t-test for unequal sample variance. The overall prevalence across all three periods of hypomagnesaemia and severe hypomagnesaemia during that episode of hypokalaemia was 70.2% and 12.9%, respectively.

Reflective testing led to a highly significant increase in number of cases of hypomagnesaemia in severe hypokalaemia diagnosed, compared to baseline (p<0.001). Subsequent introduction of reflex testing led to a further significant increase in diagnosis of hypomagnesaemia (p = 0.001). Increases were seen in diagnosis of severe hypomagnesaemia but these did not achieve statistical significance, except with reflex testing compared to baseline (p = 0.038) (fig 3). The “number needed to diagnose” (NND), that is, the average number of tests needed to be performed to diagnose a single case,4 was calculated for all three periods. The NND for hypomagnesaemia was 1.48 and 1.42 with reflective and reflex testing, respectively, and the NND for severe hypomagnesaemia was 7.67 and 8.22, respectively, implying little difference in the specificity between the two methods. Baseline NNDs were 1 and 4 for hypomagnesaemia and severe hypomagnesaemia, respectively, although these figures are likely to be influenced by the small number of cases.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Diagnosis of hypomagnesaemia. *p<0.001 compared to previous system; **p<0.01 compared to previous system.

Discussion

The overall prevalence of hypomagnesaemia (70.2%) in our data, which consisted only of severely hypokalaemic samples, is significantly higher than published rates (39–42%) in hypokalaemia per se,5 6 implying a relatively greater need to measure serum magnesium concentration in severe hypokalaemia. However, prior to reflective testing, we only diagnosed hypomagnesaemia in 7.7% cases of severe hypokalaemia, reflecting the fact that serum magnesium concentration was rarely measured in this situation.

Reflective testing significantly increased our diagnosis of hypomagnesaemia compared to the previous system. Interestingly, requesting of serum magnesium concentration increased from clinicians as well as laboratory staff, perhaps reflecting greater awareness among clinicians of the association after a number of cases had been diagnosed by the laboratory.

However, implementing reflex testing led to further increases in diagnosis of hypomagnesaemia. As both reflex and reflective had similar NNDs, they both have a similar specificity. The reflex rule however resulted in a standardised approach such that more serum magnesium levels were requested overall in this population and thus the increased absolute detection of hypomagnesaemia occurred due to this. The absolute number of extra magnesium tests requested due to the reflex testing compared to the reflective testing was just 28 tests in the 6-month period. While we did not formally assess the clinical impact of providing a diagnosis of hypomagnesaemia when it may not have been made, when one compares the cost of less than 30 magnesium tests to that of discharging one patient one day later due to delayed detection of the aetiology and correction of hypokalaemia, it is clear that the potential savings and improved quality are not insignificant.

We are aware of only one study which compares reflex and reflective testing.4 Here, the authors compared their own reflective practice of screening for haemochromatosis using iron studies with a published study in which all patients with a raised alanine transferase were screened. They found that the reflective method had a significantly lower NND. They acknowledge that the NND will be influenced by thresholds used for reflex testing, and the individual reporting practice in reflective testing. The selection of such thresholds ideally should be evidence-based but unfortunately this evidence rarely exists. Each threshold has its own sensitivity and specificity of predicting abnormalities in the added-on test, and the importance of identifying (or missing) such an abnormality will determine the exact threshold to be used. We chose a potassium concentration of 2.5 mmol/l, as severe hypokalaemia has a higher risk of complications, and may not be treated effectively if hypomagnesaemia is not also recognised and corrected. Undoubtedly additional cases of hypomagnesaemia would have been identified if looked for in less severe hypokalaemia, but these patients would have a lower risk of complications. In addition, the NND would increase significantly as far more serum magnesium analyses would be performed.

Reflective testing implicitly depends on thresholds too, it is just that the threshold in each case changes depending on other information available. In this case, additional information which may affect the decision to add on magnesium might include known hypomagnesaemia in the past, clinical information indicating possible magnesium depletion, for example, gastrointestinal losses, malnutrition, alcohol abuse, renal wasting caused by primary renal disease, hyperaldosteronism, diuretic use, osmotic dieresis in diabetes mellitus, and so forth. It should be acknowledged that in these situations there is a similar lack of evidence for what threshold should be used.

There are additional considerations which may affect whether reflex or reflective testing is more appropriate. In the study quoted, reflective adding on of tests was performed by a consultant chemical pathologist. There will inevitably be differences in the practice of different staff in the same laboratory depending on their training and individual experience, which could lead to a non-standardised service. In addition, pressures such as increased workload may also lead to staff working differently. Reflex adding on of a test is likely to be quicker than if done manually, which may be important for critical homoeostatic derangement such as severe hypokalaemia and hypomagnesaemia, where rapid action is required to prevent complications. If an appropriate threshold has been set, a rule based system is likely to be more consistent in applying the specified criteria than human intervention, be that within or outside the laboratory.

We have not addressed the “acceptability” to clinicians and the public of reflex versus reflective testing of magnesium, but it seems likely this would not be an issue; more “sensitive” tests (in the emotive sense) would require extra consideration, however.7 8 A further consideration is the relative cost of each approach. Where the NND is significantly lower for reflective than reflex testing, the cost of manual reflective testing by laboratory staff may be higher than that of the additional “negative” tests performed and the cost of the time to write the logic rules). As there are approximately 75 patients presenting with severe hypokalaemia every six months at this hospital, the reflex addition of magnesium is cost effective. Of course, this may be reversed for more expensive tests.

It should be noted that the two approaches are not mutually exclusive. For example, reflex testing with a low threshold may be used, with reflective testing for the remainder when appropriate. Thus, the consultant staff can still add serum magnesium to the request when the serum potassium concentration is >2.5 mmol/l. Such a scenario results in the best of both worlds.

It is worth noting that in common with most clinical laboratories, and the original studies on which prevalence estimates are based, our analysers measure total serum magnesium as opposed to the ionised form. It is recognised that total magnesium is not a perfect indicator of magnesium status; for example, it is susceptible to changes in albumin concentration. Seeing as albumin was not measured on all our samples, we were unable to determine any potential effect of hypoalbuminaemia or hyperalbuminaemia. However, the proportion of albumin-bound magnesium is significantly lower than that for calcium (for which serum values are routinely adjusted for albumin concentration), and such “corrected” magnesium concentrations are not frequently reported by clinical laboratories. There are other issues surrounding the direct measurement of ionised magnesium too, such as compatibility of results obtained by different direct electrodes.9 As an additional weakness of the study, we were not able to assess any possible seasonal variation in request patterns as a reflection of seasonal variation in admissions, as each test period lasted 6 months only. It might be expected however that this would affect the absolute number of hypokalaemic specimens rather than the proportion in which magnesium concentration was measured.

We conclude that testing for hypomagnesaemia in severe hypokalaemia has been more effectively achieved in our laboratory by reflex testing than by reflective testing, due to the apparent increased sensitivity, similar specificity, higher speed and lower cost. Laboratories need to consider these and other factors whenever post-analytical adding-on of tests is being contemplated.