Continuous variables, such as blood pressure and fasting blood glucose measurements, are very common in clinical research. When describing the central tendency and variability of continuous variables, data following a normal distribution are typically described using the mean ± standard deviation, while skewed data are often described using the median (interquartile range) or the median (the first quartile ~ the third quartile).

n general, when the effect size is reported as the median in the literature, it indicates that the authors are aware of the importance of normal distribution in descriptive statistics for continuous variables and have used appropriate methods based on the data distribution. If only the median, quartiles, or extreme values are reported, it suggests that the data in that study is likely not symmetrical.

On this webpage, the method for calculating based on the median and interquartile range in Scenario 1 is based on the theory of a completely normal distribution.

According to the literature, the most common scenarios are as follows: Scenario 2 = {Median, First Quartile, Third Quartile, Sample Size} Scenario 3 = {Median, Minimum, Maximum, Sample Size} Scenario 4 = {Median, First Quartile, Third Quartile, Minimum, Maximum, Sample Size}

There are four methods for converting data in Scenarios 2, 3, and 4. The estimates proposed by Tong et al. are based on the assumption of normality, click here to view the latest algorithm maintained by the authors. To address approximate conversion for skewed distributions, McGrath et al. (2020) proposed a method based on the Box-Cox transformation and quantile estimation (QE). Cai et al. (2021) proposed an estimation method based on the unknown MLN method, click here access the R shiny program。

When many of the included studies report descriptive information such as medians, a meta-analysis based on the median as the effect size can be performed. click here to using the R package 'metamedian'.

Update History:

• 2024-04-10: Tong et al.'s method updated with normality test [1]; BC, QE, and MLN methods based on estmeansd (version 1.0.1, released: 2023-12-14).
• The BC method used on this webpage does not employ Monte Carlo simulation, ensuring consistent results with each run