Analysis
First, payment data were descriptively analyzed. Payments per physician
were also calculated only for physicians receiving payment each year, as
in other previous studies.[7, 12, 14, 32] Second, the payment
concentration was evaluated by the shares of the payment values held by
the top 1%, 5%, 10%, and 25% of the otorhinolaryngologists and the
Gini coefficient at the physician level. The Gini index ranges from 0
to1, and the greater the Gini index, the greater the disparity in the
distribution of payments.[4, 7, 33] Third, we calculated descriptive
statistics and evaluated payment differences among the leading
otorhinolaryngologists, including guideline authors, society board
members, journal editors, and other otorhinolaryngologists. The
differences in payments by each variable were evaluated by Chi-square
and fisher exact tests for the proportion of otorhinolaryngologists
receiving payments and by Mann-Whitney U test for payment values per
otorhinolaryngologist. Furthermore, the linear log-linked Poisson
regression model was used to assess the association between relative
risk of payment receipt and the otorhinolaryngologist characteristics.
To account for the skewed distribution of payment values, negative
binomial regression model was employed to evaluate the association
between relative monetary value of payments per physician and the
otorhinolaryngologist characteristics. Finally, we evaluated the trends
in payments per physician and number of physicians receiving payments
between 2016 and 2019 by the population-averaged generalized estimating
equation (GEE) with the panel data of the annual payments. As the
payment distribution was highly skewed (Supplemental Material 1), the
negative binomial GEE model for the payment values per physician and
linear log-linked GEE model with Poisson distribution for the number of
otorhinolaryngologists with payments were selected.[7, 34] The
payment values were converted from Japanese yen (¥) to US dollars ($)
using the 2019 average monthly
exchange rates of ¥109.0 per $1. All analyses were conducted using
Microsoft Excel, version 16.0 (Microsoft Corp) and Stata version 15
(StataCorp).