Page 57 - Clinical Small Animal Internal Medicine

3 Using Data for Clinical Decision Making 25

Studies where individuals serve as their own controls Conclusion
VetBooks.ir lead to collection of dependent data, in which measure- Using data to make causal inferences is a hallmark of evi-
ments from the same individual are correlated. This
arises, for example, when readings are taken from both
eyes, when blood parameters are sequentially monitored, dence‐based medicine, but no inferences will be correct
if the evidence used to establish them is flawed. Research
or when drug concentrations are evaluated for half‐lives. can generate new knowledge, but it can also embed
Conventional statistical procedures assume that data are flawed conclusions in the literature if sedulous attention
independent, but when data are paired or repeatedly col- is not afforded to proper study design and conduct. Even
lected from the same individual, the independence the most sophisticated statistical methods cannot sup-
assumption is violated, which leads to invalid hypothesis plant biases that insidiously enter studies through flaws
test results, typically erroneously low P‐values and higher in study design. Despite the zeal that investigators have
than planned type I error proportions. Analysis of for finding statistically significant study results, reported
dependent data typically requires specialized statistical point and variance estimates, confidence intervals, and
methods, from Student’s t‐tests or the nonparametric P‐values will be incorrect unless great care is exercised in
Wilcoxon signed‐rank tests for paired data, to one‐way designing and establishing the central features of a clini-
repeated measures analysis of variance or the nonpara- cal study: thoughtful selection of comparison groups,
metric Friedman test for sequential observations, to mul- unbiased and masked collection of information, diligent
tivariate models, such as mixed effects analysis of variance follow‐up of patients, proper specification of statistical
or linear regression models for more complex designs. models, and verification of the model assumptions.

References

1 Buhles W, Kass PH. Understanding and evaluating 3 Hulley SB, Cummings SR, Browner WS, Grady
veterinary clinical research. J Am Anim Hosp Assoc DG, Newman T. Designing Clinical Research,
2012; 48(5): 285–98. 4th edn. Philadelphia, PA: Lippincott Williams &
2 Lecouteur RA. It’s time. Vet Surg 2007; 36(5): 390–5. Wilkins, 2013.