๐ ็ธฝ็ฎ้ ๏ฝ ๐ ่ฑๆๅๆ๏ผๆฌ็ฏ๏ผ ๏ฝ ๐ ๅฎๆด็ฟป่ญฏ ๏ฝ โญ ็ฒพ่ฏ็ญ่จ
Statistical methodology
Statistical methodology
Understanding basic statistical concepts and tools can help reduce bias from potential confounding variables and improve understanding of the application of clinical findings. Statistical analysis can be performed using one of many analysis sources (SAS, STATA), the use of which is beyond the scope of this chapter. Quantitative data are those that will be collected in most observational and experimental studies. Continuous
data are measured in real numbers with intermediate values (such as a height of 62.5 cm). Discrete data are those that are real numbers without intermediate values (number of family members with a history of skin cancer).
There are various ways to identify the center of the data. Mean is the average value of the data, including the outliers, which may greatly skew the mean. The median is the value directly in the middle (or the 50th percentile) of all the data points numbered from the lowest to the highest. Outliers do not significantly affect the median. The mode is the most common value in the data-set (or the peak of a histogram). The spread of data values can be described by variance, which is the average of the squared difference between a data point and the mean. Standard deviation is the square root of the variance, and measures how close data falls from the mean. For instance, a low standard deviation means most values fall close to the mean, while a high standard deviation suggests data falls over a wide range of values. Most data will fall in a bellshaped curve in a normal (or Gaussian) distribution. This assumes that 68% of data will fall within one standard deviation on either side of the mean, 95% will fall within two standard deviations of the mean, and 99.7% of data points will be within three standard deviations of the mean.
The p-value is a measure of the likelihood that the result could be seen by chance. Overall, a p-value <0.05, or less than 5%, is considered the threshold for being statistically significant; however, the lower the p-value, the more likely the result is to be true. Power reflects the number of data points or study participants needed in the study to reveal a statistically significant (p < 0.05) difference in outcomes between two groups. As the power increases, there is less likelihood of a false negative result since the power is 1 minus the false negative rate. In general, power >80% is considered to be statistically powerful. Prior to initiating a study, a power analysis should be performed in order to determine the minimal number of data points required to reveal a significant result. The basic information needed to calculate the sample size includes the test type (one- or two-sided), significance level (usually 0.05), desired power (1-b, or type II error, usually 0.8), and effect size estimate (an estimate of group differences based on prior clinical studies or experience). Statisticians, statistical software, and online power calculators may be used to generate sample size numbers. Without this analysis, there may be too few data points to reach an accurate conclusion, or there may be far too many participants enrolled, which poses an ethical hazard of putting patients in unnecessary harm.
Parametric tests are those used to compare means when the data are normally distributed. The most well-known parametric test is a t-test, which can be applied to continuous data with a normal distribution. The t-test used will depend on whether the groups have equal or unequal variances. A paired t-test may be used when samples are correlated, for instance, in a before-and-after study or when samples are matched pairs such as in a case-control study. Nonparametric tests are those that are used when the
data-set is not normal and include the Wilcoxon MannโWhitney U-test, KruskalโWallis (H) test, and Wilcoxon Signed-Rank test. A chi-squared test can be used when comparing an observed proportion compared to an expected proportion, such as comparing the incidence of nodular melanomas between sexes when compared to the expected value (50:50).
A regression analysis allows the researcher to test the effects of multiple independent variables on a given outcome. For instance, instead of simply determining the effect of age on metastatic rate of SCC, regression analysis could determine the effect of age, immune status, therapy received, and comorbidities on the metastatic rate of SCC in that study group. Regression analysis is performed using a statistical package that models the data resulting in an r2-value. This value can help the reader or researcher determine how much of the variance in the outcome is due to the variables identified. For instance, if r2 = 0.78, 78% of the outcome is due to the defined variables.
Logistic regression seeks to model the probability of an event occurring when the dependent variable is binary and generates an OR and p-value for any given independent variable.