Pollard JD, Deyhim A, Rigby RB, Dau N, King C, Fallat LM, Bir C.  Comparison of pullout strength between 3.5mm fully thread, bicortical screws and 4.0mm partially threaded, cancellous screws in the fixation of medial malleolar fractures.  J Foot Ankle Surg.  2010 May-Jun; 49(3): 248-52.

WHY did the authors undertake this study?
Foot and ankle surgeons have several options with respect to internal fixation of displaced medial malleolar fractures.  The aim of this study was compare one specific component of hardware failure between two common internal fixation techniques. 

HOW did they attempt to answer this question?
The primary outcome measure of the study was a direct comparison of screw pullout strength between two internal fixation constructs:  (1)  2 x 3.5mm fully threaded bicortical screws (70-75mm in length) and (2) 2 x 4.0mm partially threaded unicortical cancellous screws (40mm in length).

The study was performed on 5 cadaveric specimens in which a Mueller type B fracture was simulated.  Screw failure was achieved and measured with an Instron 8500 Plus machine, and failure was defined as 2mm of distraction. 

WHAT were the specific results?
A statistically significant difference was found in the median force required to achieve 2mm of displacement between the two groups (116.2 [range 70.2-355.5] Newtons versus 327.6 [range 117.5-804.3] Newtons; p=0.043).

HOW did the authors interpret these results?
From these results, the authors conclude that 3.5mm bicortical screws provide a more stable construct for fixation of medial malleolar fractures compared to 4.0mm unicortical screws IF one is worried about screw pullout strength as a primary cause of fixation failure.

House AA, Eliasziw M, Cattran DC, Churchill DN, Oliver MJ, Fine A, Dresser GK, Spence JD.  Effect of B-vitamin therapy on progression of diabetic nephropathy: a randomized controlled trial.  JAMA.  2010 Apr; 303(16): 1603-9. (Pubmed ID#: 20424250)

WHY did the authors undertake this study?
Elevated levels of plasma homocysteine have previously been associated with nephropathy, retinopathy, neuropathy and other vascular complications in diabetic patients.  At the same time, B-vitamin therapy has been proposed and demonstrated to decrease plasma homocysteine levels.  Ergo, the authors of this study aimed to demonstrate the effect of B-vitamin therapy on the progression of diabetic nephropathy.

HOW did they attempt to answer this question?
The primary outcome measure of this study was progression of diabetic nephropathy, as measured by the glomerular filtration rate (GFR). Through a multicenter, double-blind and placebo-controlled study design, patients were randomized to receive either placebo or a vitamin B tablets (containing 2.5mg/d of folic acid, 25mg/d of vitamin B6, and 1mg/d of vitamin B12).

The population cohort consisted of a group of 252 diabetic patients already carrying the diagnosis of diabetic nephropathy (but without advanced renal failure).

WHAT were the specific results?
As hypothesized, those patients in the B-vitamin group demonstrated statistically significant lower values of plasma homocysteine (p<0.001).  Unexpectedly however, those patients in the B-vitamin group had a more rapid decrease in renal function, and were more likely to suffer from a vascular complication (MI, stroke, revascularization or all-cause mortality).

HOW did the authors interpret these results?
Based on these results, the authors caution  practitioners with respect to the use of B-vitamin therapy in diabetic patients for the purpose of its homocysteine-lowering effect.

Let’s take a closer look at the topic of confidence intervals, specifically as they apply to the House et al diabetic nephropathy study.  There is a free link to this article, so I encourage everyone to open the .pdf file and refer to Tables 2 and 3 (on page 1607) as we go through this discussion.

There are many different ways to describe data, and these are usually referred to as descriptive statistics.  These provide characteristic information about a certain group, but do not typically describe any differences between groups.  Common examples include the mean, median, mode, standard deviation, range, etc.  For a normally distributed population, it is most correct to describe data using the mean and standard deviation.  For a non-normally distributed population, it is most correct to describe the data with the median and interquartile range.  Another way to describe data from a normally distributed population is with a confidence interval.  Instead of reporting the mean and the standard deviation, it is sometimes more useful to report the mean with a confidence interval.  A confidence interval can be any range (90%, 95%, 99%, etc.), but it is most common to report the 95% confidence interval about a mean.  As compared to the standard deviation, confidence intervals do a better job at describing the precision of the mean. 

Confidence intervals may also be useful when used with comparative statistics, or when comparing characteristic data between groups.  The first row of Table 2 presents data with respect to the “Radionuclide GFR” between the placebo and B-vitamin groups.  In the second column from the right, we see that the mean difference between the groups was -5.8 with a 95% confidence interval ranging from -10.6 to -1.1.   In the very last column we see that a p-value of 0.02 is reported, indicating statistical significance in this mean difference (assuming a level of significance of 0.05).   However, we almost don’t need the p-value reported because we can already assume that there is a statistically significant difference just based on the confidence interval alone!  When comparing means (as in this example), if the confidence interval of the mean difference does not cross “0” in the range, then the data is almost certainly statistically significant.  On the other hand, if the confidence interval of the mean difference does cross “0” in the range, then the data is almost certainly not statistically significant.

Take a minute to go through the rest of Table 2 and see if this rule of thumb works out.  The 95% confidence interval of 3 of the rows (Radionuclide GFR, MDRD GFR and Plasma total homocysteine) do not cross “0” and report p-values less than 0.05, while the 95% confidence intervals of 2 of the rows (Proteinuria and MMSE score) do cross “0” and report p-values greater than 0.05.

Table 3 however throws a curveball at us and we have to be careful using this rule of thumb.  In this table the second column from the right reports on the hazard ratio with a 95% confidence interval, and the column all the way to the right presents the p-value.   Here none of the reported confidence interval ranges cross “0”, but only two of the p-values indicate statistical significance.  So what’s going on?

Hazard ratio are similar to odds-ratios (like we’ve talked about in a previous journal club) in that a critical reader is attempting to determine if one group is as likely to have some type of outcome relative to another group.  Using the myocardial infarction row from Table 3 as an example, we can say that the B-vitamin group was 2.1 times as likely to suffer a MI than the placebo group (although this was not statistically significant).  Here our magic number for looking at the range of the confidence interval is not “0”, but instead “1”.  In other words, a hazard ration of “1” indicates that the two groups are as likely to have some unfavorable event occurring at any point of time. 

So let’s take a second look at Table 3 using “1” as our reference number when examining the confidence intervals.  The confidence interval ranges of 6 of the rows cross “1” and report p-values greater than 0.05.  Two of the ranges do not cross “1” (although they do include it) and report p-values less than 0.05.  It is a little interesting that “1.0” is reported in the confidence interval range, and one can easily appreciate how close those p-values come to 0.05. 

This information is a little heavy, but it does provide a quick means for a critical reader to quickly “check” the statistical significance of a reported variable.