"Statistics - the shortest route between an unfounded hypothesis and a foregone conclusion"

Many articles have been published on the relationship between blood lactate and outcome. It does seem likely that failure to clear lactate is associated with a high mortality. Unfortunately, the studies examining this issue are sometimes less than perfect.

Take the second article. Two hypotheses are proposed, the first that "..elevation of serum lactate levels, in the absence of clinical signs of shock, correlates with an increased incidence of death, MSOF [multiple system organ failure], and RC [respiratory complications] after severe trauma. The second is that "early identification and correction of [occult hypoperfusion] improves survival and decreases the incidence of MSOF and RC that follows severe trauma".

Unfortunately neither hypothesis is substantiated by the article! Many small criticisms can be levelled at the study. The "pilot study" was merely a retrospective chart review. The study group was heterogeneous. Hypotension was defined as a systolic blood pressure of 100, but in patients with a mean age of ~ 45 ± ~ 20 years, there could have been a substantial number of patients who had baseline hypertension, and were thus under-resuscitated at such levels; likewise, patients with extremely low diastolic pressures may not have had adequate organ perfusion. It is not clear whether anyone received adrenaline, which can result in high lactate levels without necessarily implying a poor prognosis. Patients with (one deduces) any one of (a) a systolic blood pressure under 100 mmHg, (b) a heart rate over 120/min, or (c) urine output under 1 mL/kg/hour, were excluded from analysis as they were considered to be "in shock". Someone with their usual baseline blood pressure of 90/60 (and a heart rate of 60/min), or a patient with a heart rate of 130/min (due to pain, with say a blood pressure of 170 systolic and a urine output of 2ml/kg/min), might have been inappropriately excluded! However all of these 'minor' criticisms pale into insignificance when regarded in the light of the study design.

There was no randomisation. Groups to be compared were defined according to how long it took blood lactate levels to become normal, but management depended on lactate levels. Swan-Ganz catheter placement was as follows:

Pulmonary Artery Catheter Placement
Group (number)	Deaths	"MSOF"	PAC Placement
1 (21)	0	0	0
2 (19)	0	0	7
3 (11)	0	0	3
4 (14)	3	0	9
5 (14)	5	6	8

In addition, the decision to allocate patients into the five groups (no lactic acidosis, or clearance of lactate by 6, 12, 18 or 24 hours) was only decided on analysis, ie post hoc . So in a very real sense, the study is retrospective !

The authors assert that the mortality rate in group five was "significantly higher than in all other groups", but we learn that "To analyse differences in outcome and categorical variables, the chi-squared test was used". There is a problem with this usage - the chi-squared test is only valid for large samples. A rule of thumb is that the test should be avoided if any cell in the table of expected values contains a number under five (unless the total for that group is over forty). Constructing our table of expected values for deaths, noting that overall, eight of seventy-nine patients died, we get:

Chi-squared expected values
Group:	1	2	3	4	5	Total
Lived	18.8	17.1	9.9	12.6	12.6	71
Died	2.1	1.9	1.1	1.4	1.4	8
Group Total	21	19	11	14	14	79

Using our above criterion, a Chi-Squared test is wholly inappropriate! (Also note that if we use our Chi-Squared test to compare multiple groups, the test does not indicate which group differs from the others, merely that there is an overall difference ). Some homework - which test should the authors have used?

Reconstructing the scene of the crime, we can conclude that the authors eyeballed the results, thought "Hmm, if we lump the patients according to how long it took the acidosis to clear then all those who died had persistently high lactate levels. Okay, let's now find some statistics to substantiate this contention" !

The study epitomises statistical mis-application. It is effectively not prospective but retrospective, as groups were chosen post-hoc . In view of the lack of randomisation into a treatment and control group, we certainly cannot support the authors' conclusion that "early identification and treatment of 'occult hypoperfusion' directly impacts on patient survival and patient morbidity". We merely observe in passing (before we consign the article to the statistical rubbish tip) that those patients most fiddled with, appear to have done worst. We should probably resist the temptation to re-analyse the autors' data. There is a point in some resuscitations where one has to simply stand back and say "the patient is too far gone to continue our attempts to revive him"!

Ed