STATISTICS Mean ± SD


Extreme value


Median and IQR 50 75 100 125 Turnaround time (minutes)


Fig 2. Worked example showing how one extreme value can influence different summary statistics. Most turnaround times cluster within a relatively narrow range, while one markedly delayed case extends the distribution. The mean and standard deviation are influenced more strongly by this extreme value than the median and interquartile range.


A common source of confusion


is that intervals are often discussed imprecisely. A confidence interval does not describe where most individual values lie. It describes uncertainty around an estimated quantity. That is very different from a reference interval, which aims to describe the distribution of values in a reference population, and different again from a clinical decision limit, which is linked to diagnosis, prognosis, or treatment decisions. These distinctions are not mere terminology. They reflect different questions being asked of the data.


Outliers bring together many of these


ideas. An outlying value is not a diagnosis in itself; it is a prompt for review (Table 2). A statistical outlier is a value that appears unusually far from the rest of the dataset according to some rule or model. But that definition says nothing about why it is unusual. An outlier may be a transcription mistake. It may be a sample from the wrong patient. It may reflect haemolysis, delayed transport, contamination, assay interference, or instrument malfunction. It may also be entirely genuine. In a clinical dataset, a rare but true observation can be the most important value in the set. Horn and colleagues, along with subsequent work evaluating outlier procedures, have shown clearly that outlier treatment is not trivial and that the chosen method maters. The correct response to an outlier is


therefore investigation, not reflex deletion. A practical review would usually include checking the raw data entry, confirming sample identification, reviewing sample quality and pre-analytical history, considering whether the value belongs to a different subgroup, and assessing whether the measurement is analytically plausible. Only after that can a sensible decision be made. Sometimes the value should remain in the analysis. Sometimes it should be excluded, but only with a clear reason. Sometimes it is best handled


separately in a sensitivity analysis. What should not happen is silent removal.


How to handle data in practice A practical approach to a new dataset can be summarised in a small number of steps. First, plot the data. Second, ask what overall shape is plausible: symmetric, skewed, mixed, truncated, or affected by one or more extreme observations. Third, choose summaries that fit both the shape of the data and the purpose of the description. Fourth, add a confidence interval around the main estimate wherever possible. Fifth, review unusual values before taking any decision about exclusion. Sixth, document what was done and why.


This is not a rigid algorithm. The aim is


not to force every dataset down the same pathway, but to make sure the descriptive choices are deliberate and explainable. In some setings the mean will still be the most useful summary even in the presence of skewness, provided the reader understands what it represents. In others, median and percentile summaries will be more appropriate. The important thing is that the choice is justified by the structure of the data and the practical question being asked, rather than by habit.


Worked example Consider a small local audit of turnaround times for a specialist coagulation assay. Thirty patient samples are reviewed. Twenty-seven of the results are reported between 46 and 79 minutes, two take 92 and 98 minutes, and one takes 210 minutes (Fig 2). A quick glance at the raw numbers already suggests asymmetry, but a simple dot plot would make the point much more clearly: most observations cluster in a fairly tight region, while one value sits far to the right. If the laboratory reports the mean


turnaround time alone, the estimate is pulled upward by the 210-minute case. The standard deviation also becomes


relatively large because it is strongly influenced by that extreme delay. The resulting summary may imply that the process is both slower and more variable than most individual cases would suggest. If the same dataset is summarised by the median and interquartile range, the centre and spread are less affected by the single extreme observation and may beter reflect what most patients experienced. This effect is illustrated in Figure 2, where one extreme delay influences the mean and standard deviation more than the median and interquartile range. At that point, however, the choice


of summary depends on the question. If the purpose is to describe the typical experience of the service, the median is probably the more informative centre. If the purpose is to estimate the mean operational burden across all requests, including rare long delays, the mean may still be relevant. Neither summary is inherently correct in isolation. The correct choice depends on what aspect of performance is being described. For a KPI the percentage of samples achieving the registered TAT is best, and an additional 95th or 99th percentile can also be informative. Now consider uncertainty. Suppose the


median turnaround time is 61 minutes. Reporting “median 61 minutes” is not enough on its own, because the estimate is based on only 30 cases. A confidence interval around the median would show whether that figure is fairly stable or whether a different sample of 30 cases might have produced a noticeably different estimate. If the interval is reasonably narrow, the laboratory can be more confident in using the figure for local discussion. If it is wide, the summary is still useful, but it should be interpreted more cautiously. This is exactly why interval estimates are beter than isolated point estimates in small service datasets. Finally, the 210-minute case needs interpretation. If review shows that the


May 2026 WWW.PATHOLOGYINPRACTICE.COM 25 150 175 200


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52