Complications in Canine Cranial Cruciate Ligament Surgery. Ron Ben-Amotz

      where attributable risk is the difference between the frequencies of a successful outcome between two interventions. The NNT assesses how many patients would have to be treated with one treatment compared to another before there would be one additional positive outcome. For example, if surgery A had a success rate of 75% and surgery B had a success rate of 67%, the NNT would be 13 (1/(0.75−0.65) = 12.5); the result of 12.5 was rounded up to the nearest whole number to make the result more clinically relevant and prevent overestimating the benefit of surgery A. This means that surgery A would have to be performed on 13 patients before one additional beneficial outcome would be seen. In human medicine, a single‐digit result is generally considered a good NNT [28].

      Conversely, NNH assesses how many patients need to be treated with one treatment compared to another before you would see one additional complication. NNH is calculated similar to NNT but with complication frequencies. When calculating NNH, a specific complication(s) needs to be selected. For example, if surgery A had an infection rate of 10% and surgery B had an infection rate of 0%, the NNH would be 10. This means that one more infection would be seen for surgery A after completing 10 surgeries. A target NNH depends on the severity of the complication being assessed. A single‐digit NNH may be acceptable for a minor complication such as incisional redness compared to a NNH of 100 for a major complication like implant infection.

      For both NNT and NNH, it is also important to evaluate the overall frequency of success, or a complication, to direct clinical decision making. For example, if the NNH is 10 but the overall frequency of infection is 90% vs 80%, a clinician may conclude neither procedure should be used.

Finally, likelihood to be helped or harmed (LHH) can be calculated (NNH:NNT). A number >1 tells us that the treatment is that many times more likely to help than to harm. Using the above example, surgery A has a LHH of 10/13 = 0.77. When the LHH is <1, the treatment is more likely to harm than to help. A LHH <1 is usually only acceptable when the complication assessed is minor or transient. The more serious the complication assessed, the higher the desired LHH.

      Other considerations when discussing complications include the effect of duration of follow‐up and unique complications to the procedure. Time after an intervention can influence success and complication rate; a longer follow‐up period may allow for an improved treatment response but also allows for a greater opportunity for an unintended outcome, like OA. In veterinary orthopedics, 12 months has been defined as the length of “long‐term” follow‐up; this terminology is reasonable but has been inconsistently followed [27].

      It is important that statistical and clinical significance is considered when designing a study and interpreting results. Statistical significance is an important scientific evaluation of a dataset, but it is important to remember that it may be of little or no value when considering the care of an individual patient [29, 30]. One consideration is that investigations that include large sample sizes can result in clinically unimportant variables gaining statistical significance.

      Although a p value is a common method to inform us of a statistical difference between groups, effect size and minimal clinically important difference (MCID) should also be investigated. Effect size helps quantify the magnitude of the difference observed and is not influenced by sample size. It is calculated by finding the difference between the means of two groups and dividing the result by the standard deviation [31]. Cohen suggested 0.2 is a small effect size, 0.5 is an average effect size, and 0.8 is a large effect size. For perspective, when there is an effect size of 0.8, 79% of the treatment group will be above the mean of the control group (a reference chart can provide this information for all effect size values).

      Minimal clinically important difference was defined as “the smallest difference in score in the domain of interest which patients perceive as beneficial and which would mandate, in the absence of troublesome side effects and excessive cost, a change in the patient's management” [32] which was later simplified to “the smallest change that is important to patients” [33]. Although many human studies use patient‐reported outcomes to define MCID, it is often compared to objective data. The major limitations of MCID include inconsistent definitions or values for MCID, cost of treatment not being taken into consideration, and MCID often varies depending on the baseline (more severely affected patients often need a smaller improvement to be pleased with the outcome). Alternative methods to put datasets into context include utilizing confidence intervals and probability. Regardless of the statistical methods used, they should be considered prior to the onset of a study so investigators remain unbiased. Unfortunately, most veterinary journals, reviewers, and investigators rely solely on the p value to determine significance, limiting thorough evaluation of different interventions.

References