Introduction

Nurse practitioner independence can be divided into three categories based on the scope of practice - full NPI, restricted NPI, or none. Currently, twenty two states allow full NP independence, sixteen have restricted NP independence, and twelve states have none (American 2018). Increasing nurse practitioner scope of practice has been gaining popularity. The scope of practice defines what a particular job title is allowed to do. This is due in large to the growing shortage of physicians. According to the Association of American Medical Colleges, “The United States will face a shortage of between 40,800 and 104,900 physicians by 2030” (Mann 2017). As we progress, there will not be enough physicians to fill the demands of the aging and growing population.

Allowing NPI will help to solve this problem and create some new advantages. Full NPI gives nurse practitioners the same scope of practice that physicians have. These two jobs do similar tasks. “Studies have found that NPs do about 80% to 90% of what physicians do” (Maverick Health). With these two types of workers doing similar tasks, it is not a long stretch for states to change their laws to allow NPI. With this in mind, it should have a small effect on health outcomes. There are only a few studies that assess the impact of NPI on health, but many discuss the effect on quality. One study suggested that people were equally or more satisfied with having a nurse practitioner than having a physicians but found no significant difference in health outcomes (Horrocks et al. 2002). Another advantage of NPI is that it can increase people’s access to care. This is due to there being more people that can do the work physicians do. It also frees up physicians time by not overseeing as many patients and having more availability. Furthermore, NPI could have an effect on the cost of healthcare. Increasing the supply of people who can complete the sames tasks as physicians leads to lower costs. This is because hospitals will save time and can see more patients. This reduces the cost per patient. The factor that works against this is patient utilization. When people have increased access to care, patients may increase the frequency of visits they make (Martsolf et al. 2015). This causes demand to increase. Depending on how much of an effect the change in supply and demand causes, prices could increase, decrease, or stay the same. There is not enough empirical evidence to suggest either way.

This paper aims to contribute to the understanding of possible positive or negative effects of NPI on health using nationally representative survey data. I compared states with full NPI, restricted NPI and no NPI with the self-reported poor physical health days. I also compared different income groups on health. I included the whole population in the dataset because NPI has an effect on all people. My analysis will focus on the effect of NPI policy on health outcomes with a focus on what affect income has.

Methods

Data

#----------------------------------------------------------------------------------------------#
#Project: Data Challenge 1                                                                     
#Author: Katie Schultz
#Program Name: 1_dataprep
#Data Used:BRFSS2016_DC1.dta
#Created:1/31/18
#Last Revised & Notes: 

#Contents: This file converts original data from Stata to R format and recodes variables for analysis
#----------------------------------------------------------------------------------------------#

#----------------------------------------------------------------------------------------------#
#Set Up and load data.
#----------------------------------------------------------------------------------------------#
#Set working directory
setwd("U:/Data Challenge NP Independence")

#Install packages
#install.packages("foreign")
#install.packages("tidyverse")

#Library packages
library(foreign)
library(tidyverse)

#Import Data from Stata format to R
library(foreign)
BRFSS16 <- read.dta("BRFSS2016_DC1.dta")

#Look at data
#View(BRFSS16)

#-----------------------------------------------------------------------------------------------#
#Rename variables that start with _.
#-----------------------------------------------------------------------------------------------#
names(BRFSS16)[names(BRFSS16) == "_state"] <- "fips"
names(BRFSS16)[names(BRFSS16) == "_ageg5yr"] <- "ageg5yr"
names(BRFSS16)[names(BRFSS16) == "_finalwt"] <- "finalwt"

#-----------------------------------------------------------------------------------------------#
#Select Relevant Variable. Put into NPI dataframe
#-----------------------------------------------------------------------------------------------#
NPI <- select(BRFSS16, fips, physhlth, finalwt, ageg5yr, income2, educa)
#View(NPI)

#-----------------------------------------------------------------------------------------------#
#Merge in state policy data
#-----------------------------------------------------------------------------------------------#
#First open it
statepol <- read.csv("Policy Data_Schultz.csv")

#Look at it
#View(statepol)

#Name the variable that contains the fips code the same as in our BRFSS file
names(statepol)[names(statepol)== "X_state"] <- "fips"

#Merge dataframe and policy data
       NPI_statepol <- left_join(NPI, statepol, by="fips")

#Remove the other dataframes to conserve memory
rm(BRFSS16, NPI, statepol)

#-----------------------------------------------------------------------------------------------#
#Recode variables and label values
#-----------------------------------------------------------------------------------------------#
#Recode physhlth variable
#table(NPI_statepol$physhlth)

#Recode 88's
NPI_statepol$physhlth <- recode(NPI_statepol$physhlth, `88` = 0L)

#Recode 99's and 77's to NA
NPI_statepol$physhlth <- as.numeric(recode(as.character(NPI_statepol$physhlth), "77" = "NA", "99" = "NA"))

#table(NPI_statepol$physhlth)

#Recode Income2 variable
NPI_statepol$income2 <- as.factor(NPI_statepol$income2)
levels(NPI_statepol$income2) <- c("Less than 10K",
                                 "10K to <15K",
                                 "15K to <20K",
                                 "20K to <25K",
                                 "25K to <35K",
                                 "35K to <50K",
                                 "50K to <75K",
                                 "75K or more",
                                 "Don't Know",
                                 "Refused")

#table(NPI_statepol$income2)

#Recode education variable
NPI_statepol$educa <- as.factor(NPI_statepol$educa)
levels(NPI_statepol$educa) <- c("No school/kindergarten",
                                "Grades 1-8",
                                "Grades 9-11",
                                "Grade 12 or GED",
                                "College 1-3 years",
                                "College 4+",
                                "Refused",
                                "Missing")

#table(NPI_statepol$educa)

#Recode Age Variable
NPI_statepol$ageg5yr <- as.factor(NPI_statepol$ageg5yr)
levels(NPI_statepol$ageg5yr) <- c("18-24",
                                  "25-29",
                                  "30-34",
                                  "35-39",
                                  "40-44",
                                  "45-49",
                                  "50-54",
                                  "55-59",
                                  "60-64",
                                  "65-69",
                                  "70-74",
                                  "75-79",
                                  "80 or older",
                                  "Refused")
#table(NPI_statepol$ageg5yr)
#-----------------------------------------------------------------------------------------------------#

#Drop observations with missing values on any of your variables by replacing my dataframe name with yours
#in the code below.
#Pay attention - if a ton of observations are dropped, you may have a problem here
NPI_statepol <- NPI_statepol %>% filter(complete.cases(.))

The data used in this analysis for the United States adult population comes from the 2016 Behavioral Risk Factor Surveillance System and the state-by-state NPI participation from the American Association of Nurse Practitioners (2018). The BRFSS is the largest health related survey in the United States. In this study, the key outcome variable is PHYSHLTH. PHYSHLTH measures the number of bad physical health days a person has had in the past thirty days. A value of zero means the person experienced no bad physical health days. The survey question is “Now thinking about your physical health, which includes physical illness and injury, for how many days during the past 30 days was your physical health not good?” This data is self-reported by individuals. NPI is the primary independent variable. It is a categorical variable containing three values. A value of 0 is for states that have no NPI. A value of 1 is for states with restricted NPI and 2 for states with full NPI. Figure 1 provides a map of NPI by state.

Figure 1: NPI Participation By State

Figure 1: NPI Participation By State

Red State = No NPI, Yellow State = Restricted NPI, Green State = Full NPI

In addition to these two primary measures, the analytic file also includes a measure of the level of education attained by an individual, EDUCA. This variable is on an interval scale. Education is an important factor in people’s health. Education levels of the population vary by state. Having more education usually leads to better health. Another variable that is included is annual household income, INCOME2, on an interval scale. NPI should primarily affect people of lower income. This is because the policy can increase access to care and possibly reduce costs. In the analysis, I compared the effect of different NPI levels for people of different income levels. Furthermore, my analysis accounts for the effect of age which is on an interval scale, AGEG5YR. Generally speaking, as people age, their health declines. NPI should expand services to help more patients. By comparing an individual’s health with the NPI policy, factoring in for age, states with full NPI should have healthier people.

After these sample factors are taken into account and eliminating responses with missing data, the analytic sample contains 460,208 observations. The table below provides a weighted summary statistic for the effect of NPI on health. All descriptive statistics and regressions apply BRFSS sampling weights finalwt. This accounts for the non-random sample structure of the BRFSS data. The table and graph below shows the effect NPI has on mean physical health.

NPI_statepol %>%
    group_by(NPI) %>%
      summarize(meanphyshlth=round(weighted.mean(physhlth, finalwt, na.rm=TRUE),2))
NPI meanphyshlth
0 3.86
1 4.07
2 3.68 
plot_data4 <- NPI_statepol %>%
  group_by(NPI)%>%
summarize(meanphyshlth=weighted.mean(physhlth, finalwt, na.rm=TRUE))


plot_data4$NPI_factor <- as.factor(plot_data4$NPI)
levels(plot_data4$NPI_factor) <- c("No NPI", "Restricted NPI", "Full NPI")

ggplot(plot_data4, aes(x=NPI_factor,y=meanphyshlth)) + geom_col(fill="darkblue", alpha=0.4) +
labs(x="",
    y= "Mean Physical Health", 
    title="Weighted Mean Health by NPI") +
  theme_minimal()+
  theme(plot.title = element_text(hjust = 0.5)) +
  theme(panel.grid.major.x = element_blank())

Empirical Strategy

To evaluate the impact of the NPI policy on physical health, I looked at the reported number of bad physical health days in states with no NPI and compare it to states with restricted NPI and then full NPI. I first began by estimating Model 1. This model is a simple regression showing that physical health is a product of different NPI levels.

Model 1: Simple Regression

\[PHYSHLTH_{i} = \beta_{0} + NPI_{i}\theta +\epsilon_{i}\]

In this model, Beta 0 is the number of bad physical health days that people have in states with no NPI policy. No NPI is my base group to be compared to. The theta symbol means there are multiple values within the variable NPI. Theta 1 is the estimated difference in days of poor physical health between people living in no NPI states and restricted NPI states. Theta 2 is then the estimated difference between people living in no NPI states and full NPI states. A negative estimate on theta would indicate that the NPI policy improves physical health.

This method attributes differences in physical health between states with varying NPI to the policy only. This may produce a biased estimate if the states differ in physical health for other reasons. The age of the population differs across states. Places that are warmer tend to have an older population. Having an older population increase the mean number of bad physical health that should not be attributed to the policy. Also, education varies across states. Some states invest more in education than others. Having more education can help people to be healthier since they are more informed. Removing education and age from the regression would cause the policy to be the reason for the differences in health. These two variable do not affect if a state has an NPI policy of not. Lastly, income varies across states. The more income people have, the healthier they are because they are able to afford healthcare. Also, states with lower income populations tend to have less healthy populations. This means people need more primary care. States with lower income populations would positively benefit from NPI. A positive bias would be created when not accounting for income in the model. To address sources of bias, I controlled for age, education, and income in Model 2. These are all interval scales with multiple levels. Age contains fourteen levels starting with age group 18 to 24 and ending with the group 80 or older. Education contains seven levels ranging from no education to four or more years of college. Both of these variables have a refused to answer option. Lastly, income has ten levels starting with people that make less than ten thousand and ends with people that make over seventy five thousand. Income has options for refusing and don’t know. Model 2 accounts for the other variables that affect health.

Model 2: Multipe Regression

\[PHYSHLTH_{i} = \beta_{0} + NPI_{i}\theta + Income_{i}\theta + Age_{i}\theta + Education_{i}\theta + \epsilon_{i}\]

Finally, I hypothesized that the more NPI a states has, it will help people of lower income the most.This is because restricted and full NPI can possibly reduce costs and increase access to care without decreasing health. Models 1 and 2 constrain the effect of NPI to be the same across all groups. I then controlled for the interaction between NPI and household income to allow the effect of the policy to depend on income.

Model 3: Multiple Regression Allowing the Policy Impact to Depend on Income

\[PHYSHLTH_{i} = \beta_{0} + NPI_{i}\theta + Income_{i}\theta + Age_{i}\theta + Education_{i}\theta + NPI_{i}\theta*Income_{i}\theta + \epsilon_{i}\]

Results

m1_ols <- lm(physhlth ~ as.factor(NPI), data=NPI_statepol)
m1_w <- lm(physhlth ~ as.factor(NPI), data = NPI_statepol, weights = finalwt)
stargazer(m1_ols, m1_w, header=FALSE, title = "Table 2 Results for Model 1 Simple Regression", column.labels = c("OLS", "Weighted"), type="html")
Table 2 Results for Model 1 Simple Regression
Dependent variable:
physhlth
OLS Weighted
(1) (2)
as.factor(NPI)1 -0.095*** 0.212***
(0.034) (0.028)
as.factor(NPI)2 -0.658*** -0.179***
(0.032) (0.034)
Constant 4.676*** 3.860***
(0.025) (0.018)
Observations 460,208 460,208
R2 0.001 0.0003
Adjusted R2 0.001 0.0003
Residual Std. Error (df = 460205) 8.920 192.336
F Statistic (df = 2; 460205) 261.251*** 64.426***
Note: p<0.1; p<0.05; p<0.01

Table 2 contains the results of the simple regression in Model 1 by OLS and weighted least squares. The OLS estimate contains the negative sign that I expected but the weighted least squares estimate does not. Since the weights do make a difference, all estimates will use the sampling weights. The simple regression weighted least squares estimate implies that states with restricted NPI have 0.212 more days of poor physical health and states with full NPI have 0.179 less poor physical health than states with no NPI. Physical health is measured in days and these estimates are small and statistically insignificant. These estimates show that there is no impact of the NPI policy on physical health.

Table 3 contains the results from Model 2 and 3. In model 2, controlling for age, education and income has little effect on the estimated relationship between NPI and physical health from the simple regression in model 1. The estimates are very similar. This means that differences in population ages, educational attainments, and income levels across states has little effect on the relationship. Omitting these variables would not be biasing the simple regression estimates.

In model 3, I added an interaction that allows the NPI policies to depend on household income. Once this interaction is added, it changes the estimates on NPI 1 and NPI 2. With everything factored in, states with restricted NPI have populations with .883 more days of poor physical health and states with full NPI have .642 more days of poor physical health than states with no NPI. These changes are statistically significant. I expected to see the largest impact to be on people of lower income. This was not correct. The interaction actually shows that people of higher income benefit more and have reduced number of poor physical health. The largest group in states with restricted NPI that are affected was people with income of $75K or more with a value of -.907. For states with full NPI, the group affected the most was people with income between $50K and less than $75K with a value of -.815. The differences between the groups are all less than one day and the standard errors are too large to conclude any differences as statistically significant. Both models suggest that NPI policies do not impact physical health overall or by income.

m2 <- lm(physhlth ~ as.factor(NPI) + as.factor(ageg5yr) + as.factor(educa) + as.factor(income2), data = NPI_statepol, weights = finalwt)
m3 <- lm(physhlth ~ as.factor(NPI) + as.factor(ageg5yr) + as.factor(educa) + as.factor(income2) + as.factor(NPI)*as.factor(income2), data = NPI_statepol, weights = finalwt)
stargazer(m2,m3, header = FALSE, title= "Table 3 Results for Models 2 and 3", column.labels = c("Model 2", "Model 3"), covariate.labels = c("NPI1", "NPI2", "Age 25 to 29", "Age 30 to 34", "Age 35-39", "Age 40 to 44", "Age 45 to 49", "Age 50 to 54", "Age 55 to 59", "Age 60 to 64", "Age 65 to 69", "Age 70 to 74", "Age 75 to 79", "Age 80 or older", "Refused", "Grades 1-8", "Grades 9-11", "Grade 12 or GED", "College 1 to 3 years", "College 4 or more years", "Refused", "Income $10K to less than $15K", "$15K to less than $20K", "$20K to less than $25K", "$25K to less than $35K", "$35K to less than $50K", "$50K to less than $75K", "$75K or more", "Don't Know", "Refused", "NPI1*Income $10K to less than $15K", "NPI2*Income $10K to less than $15K", "NPI1*$15K to less than $20K", "NPI2*$15K to less than $20K", "NPI1*$20K to less than $25K", "NPI2*$20K to less than $25K", "NPI1*$25K to less than $35K", "NPI2*$25K to less than $35K", "NPI1*$35K to less than $50K", "NPI2*$35K to less than $50K","NPI1*$50K to less than $75K", "NPI2*$50K to less than $75K", "NPI1*$75K or more", "NPI2*$75K or more", "NPI1*Don't Know", "NPI2*Don't Know", "NPI1*Refused", "NPI2*Refused"),type="html")
Table 3 Results for Models 2 and 3
Dependent variable:
physhlth
Model 2 Model 3
(1) (2)
NPI1 0.234*** 0.883***
(0.027) (0.123)
NPI2 0.010 0.642***
(0.032) (0.165)
Age 25 to 29 0.763*** 0.760***
(0.054) (0.054)
Age 30 to 34 1.222*** 1.221***
(0.052) (0.052)
Age 35-39 1.665*** 1.659***
(0.054) (0.054)
Age 40 to 44 2.160*** 2.156***
(0.055) (0.055)
Age 45 to 49 2.902*** 2.901***
(0.056) (0.056)
Age 50 to 54 3.527*** 3.523***
(0.053) (0.053)
Age 55 to 59 3.939*** 3.935***
(0.054) (0.054)
Age 60 to 64 4.222*** 4.220***
(0.054) (0.054)
Age 65 to 69 3.488*** 3.482***
(0.058) (0.058)
Age 70 to 74 3.395*** 3.390***
(0.064) (0.063)
Age 75 to 79 3.638*** 3.632***
(0.070) (0.070)
Age 80 or older 3.954*** 3.945***
(0.067) (0.067)
Refused 1.898*** 1.897***
(0.117) (0.117)
Grades 1-8 -0.714*** -0.714***
(0.229) (0.229)
Grades 9-11 0.047 0.033
(0.226) (0.226)
Grade 12 or GED -1.077*** -1.094***
(0.224) (0.224)
College 1 to 3 years -0.901*** -0.919***
(0.224) (0.224)
College 4 or more years -1.826*** -1.843***
(0.224) (0.224)
Refused -0.897*** -0.916***
(0.289) (0.289)
Income 10K to less than 15K -0.438*** -0.443***
(0.080) (0.111)
15K to less than 20K -1.808*** -1.604***
(0.072) (0.101)
20K to less than 25K -2.884*** -2.741***
(0.070) (0.098)
25K to less than 35K -3.643*** -3.348***
(0.069) (0.096)
35K to less than 50K -4.411*** -4.093***
(0.066) (0.092)
50K to less than 75K -4.908*** -4.495***
(0.066) (0.092)
75K or more -5.567*** -5.153***
(0.062) (0.085)
Don’t Know -2.733*** -2.397***
(0.070) (0.097)
Refused -4.729*** -4.368***
(0.072) (0.097)
NPI1*Income 10K to less than 15K -0.104
(0.179)
NPI2*Income 10K to less than 15K 0.217
(0.235)
NPI1*15K to less than 20K -0.310*
(0.160)
NPI2*15K to less than 20K -0.730***
(0.211)
NPI1*20K to less than 25K -0.282*
(0.155)
NPI2*20K to less than 25K -0.431**
(0.203)
NPI1*25K to less than 35K -0.594***
(0.152)
NPI2*25K to less than 35K -0.671***
(0.198)
NPI1*35K to less than 50K -0.709***
(0.146)
NPI2*35K to less than 50K -0.578***
(0.190)
NPI1*50K to less than 75K -0.843***
(0.144)
NPI2*50K to less than 75K -0.815***
(0.187)
NPI1*75K or more -0.907***
(0.133)
NPI2*75K or more -0.735***
(0.175)
NPI1*Don’t Know -0.668***
(0.156)
NPI2*Don’t Know -0.756***
(0.199)
NPI1*Refused -0.737***
(0.157)
NPI2*Refused -0.792***
(0.203)
Constant 6.471*** 6.184***
(0.230) (0.235)
Observations 460,208 460,208
R2 0.075 0.075
Adjusted R2 0.075 0.075
Residual Std. Error 185.019 (df = 460177) 184.995 (df = 460159)
F Statistic 1,242.867*** (df = 30; 460177) 779.843*** (df = 48; 460159)
Note: p<0.1; p<0.05; p<0.01

Conclusion

This study was conducted to examine the potential impact of NPI policies. NPI is a program that is decided by state that expands the scope of practice for nurse practitioners. This program can give nurse practitioners restricted or full independence to do the same work at physicians. Prior research has suggested that this policy has no effect on health but has the potential to decrease healthcare costs, improve access to care, and not affect quality of care. My finding also suggest that NPI has no impact on health. In addition, I hypothesized that this policy would have different affects across income groups but found no statistically significant evidence. There have not been enough studies to determine the effect NPI has on health. More research is needed because of the limitations in my research. One limitation is that the BRFSS data only used cross sectional data to estimate the impact of the policy. This means that any pre-existing differences that affect physical health may be misattributed to the NPI policy. Secondly, the BRFSS data used only looks at the physical health of individuals in 2016. It does not account for states that have switched their policy.

Works Cited

American Association of Nurse Practitioners. 2018. “State Practice Environment”. Accessed February 27, 2018. https://www.aanp.org/legislation-regulation/state-legislation/state-practice-environment

Horrocks, Anderson, Salisbury. 2002. “Systematic review of whether nurse practitioners working in primary care can provide equivalent care to doctors”. Last Modified April 6, 2002. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC100791/

Mann, Sarah. 2017. “Research Shows Shortage of More than 100,000 Doctors by 2030”. Last Modified March 17, 2002. https://news.aamc.org/medical-education/article/new-aamc-research-reaffirms-looming-physician-shor/

Martsolf, Auerbach, Arifkhanova. 2015. “The Impact of Full Practice Authority for Nurse Practitioners and Other Advanced Practice Registered Nurses in Ohio”. Accessed February 27, 2018. https://www.rand.org/content/dam/rand/pubs/research_reports/RR800/RR848/RAND_RR848.pdf

Maverick Health. n.d. “What is the difference between a physician and a nurse practitioner?” Accessed February 27, 2018. http://www.maverickhealth.com/faq/nurse/582/