Sunday, February 17, 2013

Health Care Worker Deficits in Indiana

I recently became ill and tried to get an urgent appointment to my insurance assigned clinic. The scheduler told me they could squeeze me in, in three weeks. When I explained I was hoping for the next few days since I was in distress, I was told I could try to get a waiver to go to the emergency room. Not wanting to clog the already over-burdened Wishard ER, I offered to see anybody who was available, since I didn't have a preferred primary care physician. In fact, the last several times I have been to the clinic in the last 5 years, I have randomly seen whomever was available, including a nurse practitioner. Apparently now I was only able to see the exact physician I saw 2 years ago, whose first available appointment was in three weeks. After spending several hours on the phone complaining to every administrator I could find, I was finally given an appointment in 4 days (2 if you don't count the weekend).

This inspired me to investigate the geography of the physician shortage in the U.S. As a baseline, I looked at the international data about physician and nurse distribution. According to OECD data, the U.S. ranks 18th of top 20 OECD countries for number of physicians per 1,000 population (see Table 1). I obtained the physician and nursing data by averaging 2005-2010, and ranked countries by physician density.

TABLE 1: HEALTH CARE WORKERS BY COUNTRY (OECD)

CountryRankPhysicians
/1,000
Nurses
/1,000
Austria14.67.4
Norway23.913.9
Switzerland33.814.7
Spain43.74.5
Sweden53.610.9
Germany63.510.5
Italy73.46.2
Denmark83.414.5
France93.38.0
Israel103.35.0
Ireland112.912.7
Belgium122.914.8
Australia132.910.0
Netherlands142.88.3
Luxembourg152.711.0
Finland162.79.4
UK172.69.7
US182.410.6
New Zealand192.49.4
Canada202.39.0

Then I examined the U.S. by state, since research indicates that physicians tend to congregate geographically unevenly, typically based on compensation rates, so there is likely to be greater physician density in some states, and greater physician deficit in other states. Similarly nursing and other non-MD health care professionals are not evenly distributed. I used American Community Survey data (2011, 3-year estimate) to obtain number of health care workers by state for three specific occupations: physicians, registered nurses, and "other health care diagnosing and treating practitioners." I divided these by the 2010 Census total population for each state. Additionally, I obtained the number of residents per state over the age of 60 years, presuming this population will tend to utilize health care services far more frequently than younger populations. I list the relevant data for each of these groups per state (Table 2).

Additionally, I ranked each state by an index formula based on these four populations. Using a Z-score for each state for each group, I added the health care workers, and subtracted the older population, creating a composite score indicating the accessibility of health care workers per population in each state. In the composite formula, it seemed that physician totals were the most important for general health care access, so doubled that number in the equation. Each state's ranking is listed in Table 2, from least available health care workers per state, to most, per population.

Indiana ranks 40th (Graph 1). A 2012 study by the Association of American Medical Colleges states that, for Indiana, "These already severe shortages are going to become even more prevalent when considering that the number of Indiana residents over age 65 will double between 2000 and 2030, the segment of the population that uses health care services the most." The study specifically identifies, as of 2007, a 5,000 physician shortage for Indiana alone.

As a check on my health care worker index, I used CDC death rates to generate a comparison for validity. I used two measures, CDC infant mortality rates from 2007-2008 by state (CDC Wonder), and death rates (age adjusted) for all causes from 1998-2010 by state (CDC Wonder). My health care worker index had a correlation of -0.46 with infant mortality, and -0.55 for all-age death rates (see Graph 2), indicating a reasonably strong relationship--as the density of health care workers increases, both infant mortality and death rates go down, and vice versa. The correlation with the composite index is greater than any of the individual measures.

GRAPH 1: PHYSICIAN DENSITY BY STATE

GRAPH 2: RATES OF DEATH VS HEALTH CARE WORKER INDEX (CDC, 1998-2010, ALL CAUSES, AGE ADJUSTED)

TABLE 2: HEALTH CARE WORKERS BY STATE

StateRankPhysicians &
surgeons
/1,000
Registered
nurses
/1,000
Other health
practitioners
/1,000
60+ PopulationZ-Index
Florida502.58.62.723%-1.96
Nevada492.16.82.417%-1.55
West Virginia482.2102.923%-1.5
Arkansas471.99.42.720%-1.48
Oklahoma4627.92.619%-1.43
Arizona452.27.62.619%-1.38
South Carolina442.48.52.620%-1.25
Alabama432.19.72.719%-1.12
Mississippi4129.52.618%-1
New Mexico412.37.42.819%-1
Indiana402.2102.618%-0.95
Idaho381.97.13.117%-0.67
Hawaii382.77.13.120%-0.67
Georgia362.17.42.616%-0.63
Louisiana362.68.52.618%-0.63
Texas352.37.32.515%-0.58
Delaware342.9112.620%-0.47
Kentucky332.610.22.819%-0.45
Tennessee322.79.12.919%-0.38
Missouri312.89.82.920%-0.28
Oregon3037.53.120%-0.25
Michigan292.98.9320%-0.18
Illinois272.89.12.718%-0.16
Ohio272.910.52.920%-0.16
Maine263.410.5322%-0.14
California252.67.12.916%-0.13
Virginia242.78.2318%-0.1
Wyoming231.37.43.616%-0.09
Montana222.49.83.421%-0.02
Wisconsin212.410.23.219%0.01
North Carolina202.79.13.119%0.07
Iowa182.111.23.620%0.14
Kansas182.710.42.918%0.14
Utah172.26.62.813%0.36
South Dakota162.211.63.419%0.45
Pennsylvania153.510.73.421%0.67
New Jersey143.48.93.219%0.69
New Hampshire132.912.13.319%0.75
North Dakota12211.33.719%0.76
Rhode Island11410.32.920%0.79
Washington102.98.53.518%0.92
Connecticut93.910.23.320%1.2
Minnesota83.110.73.518%1.28
New York7493.419%1.31
Colorado62.88.33.716%1.39
Nebraska52.710.73.818%1.41
Maryland449.93.218%1.64
Massachusetts34.3113.319%1.77
Vermont23.311420%1.85
Alaska12.58.73.210%1.98

Tuesday, February 12, 2013

Gun Ownership and Firearm Homicide Rates

While tracking down a Facebook meme from MoveOn.org about the relationship between gun ownership and firearm murder rates, I fitfully realized that we don't have gun ownership data, thanks to legislation that forbids tracking of firearms. Research that claims to look at the relationship between gun ownership and crime base their analysis on 2001 data from the CDC's Behavioral Risk Factor Surveillance System survey from 2001, with a sample size of 201,881. Presuming people told the truth when asked if they have guns, and the CDC was able to contact a representative sample of people, they provide a margin of error rate averaging 1.8 (range 1-2.9)--however, given the nature of this population, I am skeptical of both of these presumptions. The CDC asked three questions--gun ownership, whether their guns were loaded, and whether their guns were kept in an unlocked place + loaded. Summary results can be found in the journal Pediatrics. Not to mention that this study is over 10 years old, and the world has become a very different place since 9/11, economic disaster, and a Black Democrat was elected to the presidency.

I found only one other possible source of data for gun ownership, a proxy measure from internet news, The Daily Beast, which, from 2011-2012, tracked the number of NICS background checks for firearms by state. The primary problem with this data is that each state can create its own policies for tracking and reporting background checks, so it is difficult to compare between state rates. For example, they mention that Kentucky mandates monthly checks for concealed weapons, creating a presumably inflated number, and indeed, the largest of background checks relative to the other states. Further, this is only a measure of background checks, which neither considers illegal weapons, nor whether the individual actually purchased the weapon. One might propose that both of these are random sources of error, equally distributed through states. However, given the vast differences between states, I doubt this can be considered equally distributed variance. There is a 0.38 correlation between the 2001 BRFSS gun ownershp study, and the 2012 Daily Beast background checks data. This is a moderate correlation, indicating that the data may have some reasonable level of consistency in predicting actual gun ownership, especially if combined into an index score.

A third source of estimation about guns is gun laws themselves, while, as above, not a measure of gun ownership as such, but one might presume that a breadth of gun safety laws would be associated with fewer guns. A 2010 report from Mayors Against Guns, Trace the Guns, tracked 10 different specific laws across all states. In the data provided below, I compiled this list into an index score from 0-10, with 10 being the maximum number of gun-safety laws, and 0 being none. There is a -0.36 correlation between gun laws and number of background checks, and -0.73 correlation between gun laws and gun ownership The first correlation is moderate, indicating that the more gun safety laws that a state has enacted, the fewer background checks that were reported. This could indicate either greater subversion of gun laws where they are more tightly regulated, or less interest in purchasing guns in states with more gun safety laws. The latter correlation is quite strong, indicating that in states with more gun safety laws, far fewer people admitted to owning guns in the 2001 CDC survey, than in states where there were fewer gun safety laws. Combined, these three measures are reasonably consistent with each other, although clearly problems exist in attempting to use these to determine actual gun prevalence rates in states.

Firearm death data is more reliable, although problems still exist. The FBI tracks national crime data as it is reported by states, and hosts that data publicly at the Bureau of Justice Statistics Web site. Critics are concerned that this data includes both suicide and self-defense data. I do not know if this assertion is true. However, regarding self-defense, Gary Kleck, a well-known pro-gun activist and researcher, indicates that multiple studies provide a range of 6-12% of firearm homicides that were likely "justifiable" homicides. A second source of firearm death data, from the CDC Web site, filters out suicide data, thus includes only "true homicide" rates, although it likely includes self-defense. However, given the relatively small representation of self-defense homicides, the impact on the data is likely marginal. In the data provided below, I used the total firearm homicides from 2006-2010, to control for specific, outlier years of high or low homicide rates. There is a 0.96 correlation between the 2011 FBI homicide rates, and the 2006-10 CDC firearm homicide rates. This indicates an incredibly high level of consistency between these two separate data sources, thus a substantial likelihood of reliability that these measures accurately represent non-suicide, firearm homicide rates.

Measuring criminally-inflicted death is one approach to the gun safety discussion. However, what seems more relevant are the total number of gun deaths, not just homicide. In terms of a direct, linear correlation, there is no clear relationship between gun ownership, background checks, or gun laws, and firearm homicide (correlations less than abs(0.15)). However, the picture changes dramatically when you combine all firearm deaths: homicide, suicide, and accidental deaths. In this case, correlations for each of the three gun-prevalence measures range from 0.25 for the number of background checks, 0.69 for gun ownership rates, and -0.68 for gun laws (Graph 1). So while there is only a small relationship between the number of background checks and firearm death per state, there is a very strong positive relationship between measured gun ownership rates and firearm death (i.e., as gun ownership goes up, so do firearm deaths), and a very strong negative relationship between gun safety laws and firearm deaths (i.e., as states enact more gun safety laws, firearm deaths go down).

Graph 1: All Firearm Deaths, 2006-2010 vs. Gun Ownership, 2001

Despite the poor bivariate linear relationship between gun-prevalence variables and firearm homicide, controlling for other variables generates a far stronger relationship. Specifically, both poverty and inequality contribute to firearm homicide rates, and controlling for these at the state level allows a strong relationship to appear (I also tested for 2010 urbanization but there seems to be little relationship at the state-level). Explaining the 4-variable relationship is complicated (see below). However, the 3-variable relationship between poverty, gun ownership and firearm homicides is easy to show visually with a scatterplot (Graph 2). Below you can see two trendlines, one for state poverty levels below 15% (blue), and the other for poverty levels above 15% (green). The negative slope of the line representing wealthier states indicates that higher gun ownership rates are related to lower levels of firearm homicide rates (35 states). The positive slope of the line for poorer states indicates that higher rates of gun ownership are related to higher rates of firearm homicide (15 states).

So while there isn't a simple bivariate linear relationship between gun ownership and firearm homicide rates, there is clearly a relationship once you control for poverty, and even moreso once you also control for levels of inequality. Looking just at firearm deaths in total, you don't even have to control for poverty and inequality to demonstrate a very strong relationship between gun ownership and firearm death. (Poverty data is from the 2011 American Community Survey, 5-year estimate, as is the inequality data. Inequality here is measured by the ratio of the number of households in the state living on less than $15,000 to the number of households living on more than $100,000.)

Graph 2: Firearm Homicides, 2006-2010 vs. Gun Ownership, 2001, controlling for poverty

Linear equation for the 4-variable interaction:
Firearm Homicides = -0.073 - 0.678(Inequality) + 1.068(Poverty) - 0.258(Gun Ownership X Poverty) + 0.042(Gun Ownership)
This equation explains 47% of the variance, p<0.000, and the Gun Ownership X Poverty interaction variable is significant at p<0.042.

StateFirearm Murder RateCDC Wonder, 2006-2010, Firearm Homicide (no suicide)CDC, All Firearm Deaths, 2006-10Background Check Rate/PopulationGun Ownership, 2001TraceTheGuns, 2010, All Gun Laws CombinedPoverty, 2011 ACS, 5-year Income Inequality: <15k:>100k, 2011 ACS, 5 year
Alabama717.1838851.7417.61.10452
Alaska0.0225352113.3181106857.809.50.243236
Arizona0.034730914514.8494431.1016.20.633314
Arkansas0.0377229085.215.2796855.3118.41.362871
California0.0327481614.38.6284221.3814.40.363118
Colorado0.0145158082.110.9783534.7412.50.440104
Connecticut0.0263010632.45.1633516.799.50.285685
Delaware0.0311804014.59.5312725.5511.20.385371
Florida4.512.2408124.5214.70.722721
Georgia0.0381915775.212.8444740.3216.50.721028
Hawaii0.0007352940.7312088.7910.20.311204
Idaho0.0108418371.212.3815655.3014.30.850145
Illinois0.0293819664.88.3793720.2813.10.483338
Indiana0.0282233193.811.2686739.1214.10.789097
Iowa0.0062376890.96.5468142.8711.90.709595
Kansas0.0255871012.710.5686442.1012.60.632461
Kentucky0.0230467853.213.25819647.7018.11.234068
Louisiana0.08868299110.119671144.1018.41.000716
Maine0.0090361451.18.6663440.5212.80.842797
Maryland0.0471077246.511223121.3890.241683
Massachusetts0.0186316431.73.5303312.6910.70.373835
Michigan0.045528126511.1412538.4715.70.755491
Minnesota0.0081070891.36.7784441.75110.443991
Mississippi0.0465116287.417.3708255.3121.61.558142
Missouri0.0460844885.313.6800241.7114.30.830348
Montana0.0070778561.615.31278657.7114.60.957503
Nebraska0.0230010952.28434738.65120.684228
Nevada0.0277674944.215.4504733.8012.90.468341
New Hampshire0.0045592710.67897130280.301071
New Jersey0.03059599635.193812.3109.40.262094
New Mexico0.029140359414.6649134.80190.940948
New York0.0229641862.75.11652181014.50.499277
North Carolina0.0351337184.612.2481141.3516.10.864843
North Dakota0.0089153050.68.71190150.7212.30.795394
Ohio0.029817113.59.5510232.4114.80.812913
Oklahoma0.03492402413.9909442.9016.31.003063
Oregon0.0104411381.410.8639839.8514.80.701096
Pennsylvania0.0370020474.210.6721834.7512.60.629322
Rhode Island0.0047483381.64.4210612.8712.80.560825
South Carolina0.0482162165.513.8632242.32171.00953
South Dakota0.0061425060.89.1997756.6013.80.890987
Tennessee0.0384494175.315.2763843.9216.91.044275
Texas0.0277976624.110.7526135.90170.61675
Utah0.0094066571.19.7782743.9311.40.425006
Vermont0.0063897760.99.3522742111.30.594292
Virginia0.025996753.510.6511935.1410.70.341807
Washington0.0117472121.88.9715733.1212.50.419157
West Virginia0.0232056132.713.91167055.4017.51.528704
Wisconsin0.0140671711.98.2843644.42120.615174
Wyoming0.0195035461.2161051759.7210.10.456652