Chapter 5 Results

In examining the total payments exchanged each year, there is an obvious difference in the scale of the 2015-2019 data and the 2020 data. While the 2015-2019 data is in the hundreds of millions, the 2020 data doesn’t reach $70 million. This could be due to a lag in reporting of payments, or it could reflect decreased transactions due to the COVID-19 pandemic, which began in the US in the first quarter of 2020. Thus, the 2020 data points should be considered with caution.

Aggregate payments, in 2020 dollars
Years Total Payments
2015 218,908,281
2016 102,052,022
2017 179,455,555
2018 179,486,553
2019 185,317,184
2020 69,956,239

5.1 Payment Characteristics

5.1.1 Type of Payment

Payment types are dominated by 3 types: consulting fees, food and beverage, and compensation for services other than consulting. These payment types are approximately 10 times more expensive than the other categories.

Consulting fees are "payment[s] that a company makes to a physician for advice and expertise about a medical product or treatment (Medicare & Medicaid Services (2022b)). However, this definition is extremely vague and is applied by the pharmaceutical companies themselves. The other two dominating categories have no direct association with medical expertise.

Table 5.1: Total Payment per Payment Type, in Millions
Nature of Payment 2015 2016 2017 2018 2019 2020
Royalty or License 12.33 6.02 5.59 8.09 9.46 26.47
Consulting Fee 37.57 16.58 35.10 38.83 33.87 14.86
Food and Beverage 73.97 32.09 63.57 58.46 51.13 14.03
Compensation for services other than consulting, including serving as faculty or as a speaker at a venue other than a continuing education program 56.13 24.19 45.10 46.20 60.83 7.97
Grant 5.58 4.30 4.33 4.55 3.36 2.50
Current or prospective ownership or investment interest 6.38 8.04 5.80 4.12 6.64 1.64
Travel and Lodging 17.89 8.81 14.56 13.75 13.64 0.95
Honoraria 3.13 0.55 1.87 1.62 1.65 0.62
Gift 0.23 0.01 0.79 1.92 2.07 0.38
Education 2.73 0.99 1.36 1.29 2.05 0.36
Compensation for serving as faculty or as a speaker for a non-accredited and noncertified continuing education program 2.89 0.09 1.07 0.54 0.49 0.13
Compensation for serving as faculty or as a speaker for an accredited or certified continuing education program 0.05 0.29 0.09 0.05 0.11 0.04
Entertainment 0.01 0.01 0.01 0.08 0.01 0.00
Charitable Contribution 0.01 0.08 0.22 0.01 0.01 0.00

Interestingly, while food and beverage garners huge amounts of funding on the aggregate level, it has the second least expensive average payment. Thus, this payment type must be extremely popular and often used by the pharmaceutical industry to build relationships with primary care physicians.

Table 5.2: Average Payment per Payment Type, in Thousands
Nature of Payment 2015 2016 2017 2018 2019 2020
Royalty or License 97.83 52.82 37.29 56.58 46.81 164.42
Current or prospective ownership or investment interest 193.27 423.07 290.03 216.73 255.57 126.37
Grant 7.63 10.00 9.23 9.40 5.66 7.58
Compensation for serving as faculty or as a speaker for an accredited or certified continuing education program 2.80 3.84 3.05 3.51 2.79 2.31
Consulting Fee 1.40 3.77 3.02 2.93 2.77 2.31
Compensation for services other than consulting, including serving as faculty or as a speaker at a venue other than a continuing education program 1.82 2.38 2.22 2.26 3.39 1.56
Compensation for serving as faculty or as a speaker for a non-accredited and noncertified continuing education program 2.16 2.95 1.91 2.15 2.33 1.55
Honoraria 2.03 2.17 2.53 2.25 2.36 1.45
Gift 0.14 0.14 0.62 1.96 2.28 0.61
Travel and Lodging 0.40 0.39 0.35 0.33 0.35 0.27
Education 0.03 0.04 0.02 0.04 0.05 0.08
Entertainment 0.08 0.10 0.11 0.55 0.13 0.07
Food and Beverage 0.02 0.02 0.02 0.02 0.02 0.02
Charitable Contribution 0.98 2.02 4.34 2.78 1.95 0.00

5.1.2 Population Density

As would be expected, ZIP codes categorized as “Extremely Dense” and “Dense” have the highest amount of both total and average payments, while ZIP codes categorized as “Extremely Sparse” and “Sparse” have the lowest amount of both total and average payments. This makes sense, as dollars would go the furthest when paid to primary care physicians serving ZIP codes with dense populations, as these physicians are seeing and having an impact on the greatest number of patients.

For each year of the study, one-way ANOVA analysis was performed to determine if there was a statistically significant difference in payments to ZIP codes of the 5 different density categories. Each ANOVA returned a p-value of less than \(\alpha < 0.05\), so it seems there is a significant difference between the density categories.

One-way ANOVA of Avg. Payment by Population Density Categories
Years P-value
2015 <0.001 *
2016 <0.001 *
2017 <0.001 *
2018 <0.001 *
2019 <0.001 *
2020 0.0414 *

5.2 Geospatial

5.2.1 ZIP Code

The first level of geospatial analysis considers payments aggregated by ZIP code. The table below shows summary statistics for the top ZIP code (in total payments) for each year. The table highlights the wide range of payments to ZIP codes, with the top ZIP codes in 2019 and 2020 receiving over 15 million dollars, while the top ZIP codes from 2015-2018 are under 6 million dollars.

2015-2020 Top ZIP Code in aggregation statistics ($)
Years Top ZIP Code (total payments) Total Payments No. of Payments No. of Physicians Avg. Payment Amount per Physician
2015 75390 5,336,835.31 679 125 7,859.85 42,694.68
2016 75390 5,112,483.24 438 86 11,672.34 59,447.48
2017 75390 3,300,777.35 775 140 4,259.07 23,576.98
2018 55404 3,111,817.94 89 18 3,964.25 172,878.77
2019 85020 18,656,875.69 609 67 30,635.26 278,460.83
2020 95966 15,826,688.30 57 13 277,661.20 1,217,438

As the figures further confirm, ZIP code may be too nuanced of a means for comparison, as there are over 40,000 ZIP codes in the USA, and the payments vary widely.

5.2.2 State

Thus, let us compare payments on a state level. For each year, the leading state in total payments was California. This makes sense due to California’s large population. As the table shows, there is less fluctuation year to year on the state level, so this analysis may prove more fruitful.

2015-2020 Top state in aggregation statistics ($)
Years Top State (total payments) Total Payments No. of Payments No. of Physicians Avg. Payment Amount per Physician
2015 California 35,954,080 367,403 20,365 97.86 1,765.48
2016 California 13,997,319.39 155,796 13,578 89.84 1,030.88
2017 California 25,073,460 324,136 19,309 77.35 1,298.54
2018 California 27,366,160 293,444 19,513 93.26 1,402.46
2019 California 25,846,585.07 255,469 17,793 101.17 1,452.63
2020 California 21,528,528.10 74,094 7,304 290.56 2,947.50

Each figure represents the log total payments to the continental US states for the listed year. In observing the changes in colors over the study period, it seems log total payments to primary care physicians in states in the Northeast decreased over time, while log total payments to primary care physicians in states in the West increased.

5.2.3 Region

To get a better sense of regional changes in log total payments to primary care physicians, states were grouped into the 4 regions identified below.

The plot of total payments to each region shows that the South received the highest aggregate amount for most of the time period. As anticipated above, primary care physicians in states in the West region saw an increase in aggregate payments, while primary care physicians in states in the Northeast saw a decrease.

For 2015-2019, average payments remained relatively constant, with higher average payments to physicians in the West and Northeast. However, in 2020, the West saw a sharp deviation from the other regions. This would indicate that there are potential outlier payments to primary care physicians in the West.

Again, for each year of the study, one-way ANOVA analysis was performed to determine if there was a statistically significant difference in payments to primary care physicians in different regions of the country. Each ANOVA returned a p-value of less than \(\alpha < 0.05\), so it seems there is a significant difference between the regions each year. Thus, with the figures above, it follows that physicians in the West and Northeast get statistically significant higher average payments than physicians in the South and Midwest.

One-way ANOVA of Avg. Payment by Regions of USA
Years P-value
2015 <0.001 *
2016 <0.001 *
2017 <0.001 *
2018 <0.001 *
2019 <0.001 *
2020 <0.001 *

5.3 Temporal

5.3.1 Quarter of Year

Payments are now aggregated on a quarterly basis to determine if there is a difference in payment activity related to temporal trends. Over the study period, there does not seem to be a large difference in total payments to primary care physicians between the quarters for any given year. Likewise, there does not seem to be great differences in average payments. However, there does appear to be an outlier in average payments during Quarter 3 of 2020.

The one-way ANOVA analysis of differences in the quarters returned mixed results. While it found significant differences in 2015, 2017, and 2018, it did not return statistically significant results for 2016, 2019, or 2020. Thus, more temporal analysis is necessary.

One-way ANOVA of Avg. Payment by Quarter of Years
Years P-value
2015 <0.001 *
2016 0.061
2017 0.0291 *
2018 0.0012 *
2019 0.241
2020 0.213

5.4 Models

5.4.0.1 Linear Modeling

Now, this study considers how all the previous factors influence each other over each given year. The table below highlights the results of the linear model for 2019. The models for the other years can be found in the Appendix.

In fitting linear models, the residuals need to be examined. Here, there is extreme right skew, with the residuals being non-constantly variant. In fact, the residuals fan out as the predicted values increase. This trend holds for all models, i.e. there is right skew in the data for each year.

Thus, a log-transformation is applied to the dependent variable: payment value. Although there is right skew in the data, the residuals show that there is constant variance around a value of 0. Thus, the log-transformation of the total payments data is successful in achieving the linearity necessary for the linear model below.

The baseline for each model is the log payment to a primary care physician whose specialty is Family Medicine, has an extremely sparse ZIP code in the Northeast, and the payment is for “Compensation for services other than consulting” during the first quarter.

This model confirms many of the previous findings. There are significant differences between the regions and density categories, while there are mixed results for quarters of the year. Interestingly, it seems that a physician’s specialty may have predictive power, as there is a significant difference in log payments to Family Medicine physicians and physicians of all the other specialties. The nature of payment results also confirm what was found previously, showing that some payments have much higher values than others.

Table 5.3: Model Results for Log Total Payments, 2019
Covariate Coefficient Std. Error Statistic P-Value
Intercept 7.063 0.413 17.092 <0.001
Specialty: Family Medicine|Adolescent Medicine 0.111 0.028 3.916 <0.001
Specialty: Family Medicine|Adult Medicine 0.129 0.009 14.101 <0.001
Specialty: Family Medicine|Geriatric Medicine 0.147 0.010 14.970 <0.001
Specialty: Family Medicine|Obesity Medicine 0.266 0.044 6.122 <0.001
Specialty: General Practice 0.319 0.002 129.471 <0.001
Specialty: Internal Medicine 0.086 0.001 75.181 <0.001
Specialty: Internal Medicine|Adolescent Medicine -0.134 0.015 -8.965 <0.001
Specialty: Internal Medicine|Geriatric Medicine 0.132 0.006 21.607 <0.001
Specialty: Internal Medicine|Obesity Medicine 0.300 0.041 7.348 <0.001
Specialty: Pediatrics 0.197 0.002 85.468 <0.001
Specialty: Pediatrics|Adolescent Medicine 0.213 0.011 20.253 <0.001
Specialty: Preventive Medicine|Public Health & General Preventive Medicine 0.574 0.021 26.803 <0.001
Nature of Payment: Compensation for services other than consulting -0.084 0.413 -0.204 0.838
Nature of Payment: Compensation for serving as faculty or as a speaker for a non-accredited program 0.151 0.417 0.361 0.718
Nature of Payment: Compensation for serving as faculty or as a speaker for an accredited program 0.484 0.433 1.116 0.265
Nature of Payment: Consulting Fee -0.264 0.413 -0.639 0.523
Nature of Payment: Current or prospective ownership or investment interest 3.351 0.444 7.550 <0.001
Nature of Payment: Education -4.807 0.413 -11.632 <0.001
Nature of Payment: Entertainment -3.366 0.425 -7.918 <0.001
Nature of Payment: Food and Beverage -4.528 0.413 -10.957 <0.001
Nature of Payment: Gift -1.513 0.414 -3.654 <0.001
Nature of Payment: Grant -0.381 0.415 -0.919 0.358
Nature of Payment: Honoraria 0.097 0.414 0.235 0.814
Nature of Payment: Royalty or License 1.180 0.417 2.827 0.005
Nature of Payment: Travel and Lodging -2.334 0.413 -5.647 <0.001
Region: South -0.001 0.001 -0.507 0.612
Region: Midwest -0.091 0.002 -52.720 <0.001
Region: West 0.030 0.002 17.031 <0.001
Density Category: Sparse 0.037 0.002 22.456 <0.001
Density Category: Normal 0.061 0.002 37.154 <0.001
Density Category: Dense 0.111 0.002 66.452 <0.001
Density Category: Extremely Dense 0.261 0.002 147.922 <0.001
Quarter 2 0.000 0.001 -0.055 0.956
Quarter 3 -0.016 0.001 -11.214 <0.001
Quarter 4 0.016 0.001 10.660 <0.001

5.4.0.2 Time-series Modeling

Let us build a time-series model on average payments to primary care physicians for each quarter during the study period.

In applying the Holt-Winters method to average payments for each quarter, the model-predicted values appear to be somewhat similar to the actual values through 2019, but there is a large difference for the 2020 values.

                   ME     RMSE      MAE       MPE     MAPE      MASE       ACF1
Training set 2.009594 13.41988 9.581064 0.3481101 13.43391 0.8763465 -0.4487789

For this model, the average difference between the fitted value and the actual value is MAPE = 13.43%. As highlighted above, the model seems to do particularly badly for the 2020 values. Thus, let us fit a new model for just the points from 2015 to 2019.

                   ME     RMSE      MAE      MPE     MAPE      MASE       ACF1
Training set 1.521215 9.433044 7.593289 0.594542 11.95973 0.9660007 -0.3993894

Without the 2020 data, the average difference between the fitted value and the actual value is MAPE = 11.96%. Thus, the 2020 data may introduce divergence from the trend present from 2015 to 2019.