August 18, 2011 INPATIENT PREVENTABLE HOSPITALIZATIONS FOR AMBULATORY CARE SENSITIVE CONDITIONS IN HARRIS COUNTY

Transcription

1 August 18, 2011 INPATIENT PREVENTABLE HOSPITALIZATIONS FOR AMBULATORY CARE SENSITIVE CONDITIONS IN HARRIS COUNTY Report Prepared for the Houston Endowment Project Sharanya Murty, Charles E. Begley, J. Michael Swint University of Texas Health Science Center at Houston School of Public Health Division of Management, Policy and Community Health

2 Texas has a large and growing population of low-income uninsured (DeNavas-Walt, Proctor and Smith, 2008). In 1990, there were 3.6 million uninsured individuals in Texas. By 2005, this figure had increased by 55% to 5.6 million, representing 25.4% of the population (U.S. Census Bureau). Health insurance coverage is a major determinant of health care access in the US, including access to primary care (Hoffman and Paradise, 2008). Therefore, sufficient access to primary care is a policy concern in Texas. Lack of health insurance and the consequent lack of access to care is particularly relevant in Harris County, Texas. Over one million of the county s population (over 30%) are uninsured, representing one fifth of the state s uninsured (U.S. Census Bureau). The uninsurance rate in Harris County varies dramatically by race/ethnicity with 49% of Hispanics, 31% of blacks, and only 13% of Non-Hispanic whites uninsured. A significant percentage of uninsured are adults between 18 and 64 years of age (72%); 29.3% are children between 0 and 17 years of age; and 3.4% are 65 and over (Texas State Data Center, 2005). The vast majority of the uninsured (71%) are considered low-income by federal standards i.e. between 0-199% of the Federal Poverty Level (FPL) (Texas Health Institute, 2006). Like many other communities, Houston/Harris County has a local healthcare safety net comprised of a loosely organized group of public and private providers. The public hospital and clinic system, funded through a local property tax, consists of three hospitals, eleven community health clinics, eight school-based clinics, one dental center, a program of healthcare for the homeless, and a specialty center for people with HIV/AIDS. Approved Gold Card patients with no insurance and income below 100% of the FPL are not charged for services; those with incomes up to 250% of the FPL are charged fees based on a sliding scale. The system serves about 300,000 individuals a year through all of its facilities. There are long waits for approval and appointments at some of the clinics, as well as crowding for hospital-based specialty inpatient and outpatient services. Some of the uninsured who cannot be seen in the system rely on other public agencies, private clinics or hospitals, and individual providers for charity care, or they go without care. Over the past ten years, a number of efforts have been made to expand the capacity and performance of the public and private resources of the local safety net. To assist the community in targeting specific groups or locations with access problems, it is important to determine if the lack of primary care is persistently high in these groups/areas. If it is then such groups/areas can be considered to be a priority for policy intervention. A relatively accessible measure that has been used to monitor primary care access is risk and rate of hospitalizations for Ambulatory Care Sensitive Conditions or ACSCs (Bindman et al, 1996). ACSCs are defined as conditions for which good outpatient care can potentially prevent the need for hospitalization, or for which early intervention can prevent complications for more severe disease (Agency for Healthcare Research and Quality, 2007). Hence, hospitalizations for ACSCs are also referred to as Preventable Hospitalizations or PHs. The Agency for Healthcare Research and Quality has identified 14 conditions in adults as ACSCs, and considers all hospitalizations for these conditions as preventable hospitalizations (AHRQ, 2007). The AHRQ classification system uses International Classification of Diseases, 2

3 Ninth Revision, Clinical Modification (ICD-9-CM) codes to identify ACSCs and PHs from hospitalization data. The 14 conditions classified as ACSCs by AHRQ are listed in Table 1. Table 1: Agency of Healthcare Research & Quality Ambulatory Care Sensitive Conditions Diabetes Short-term Complications Perforated appendix Diabetes Long-term Complications Chronic Obstructive Pulmonary Disease Hypertension Congestive Heart Failure Low Birth Weight Dehydration Bacterial Pneumonia Urinary Tract Infection Angina Uncontrolled Diabetes Adult Asthma Lower-extremity Amputation in Diabetes Having health insurance coverage has consistently been associated with lower risk of PHs (Chang, Mirvis and Walter, 2008; Cousineau, Stevens and Pickering, 2008; Weissman, Gatsonis and Epstein, 1992). Weissman et al observed the effect of lack on health insurance coverage on PHs in Maryland and Massachusetts and found that the odds of having a PH were 29% to 76% higher in the uninsured as compared to insured individuals. Several other social and demographic characteristics have been found to be significantly associated with PHs. The risk of PHs increases with age (Pappas, Hadden, Kozak and Fisher, 1997) and males have a higher rate of PHs than females (Culler, Parchman and Przybylski, 1998). Also, several studies have indicated that non-whites, specifically Hispanics and non- Hispanic Blacks have higher rates of PHs than Whites (Cable, 2002; Culler et al, 1998; Laditka and Laditka, 2006). PH rates are highly associated with income, with low-income groups being highly vulnerable to PHs (Cable, 2002; DeLia, 2003). PH rates have also been shown to vary with geographical location, with a small proportion of ZIP codes consistently showing significantly higher PH rates when compared to others (DeLia, 2003). These findings indicate that there may be certain groups and geographic regions that face higher risk of being hospitalized for conditions that can be managed in the ambulatory setting with appropriate access to primary care. By identifying these high-risk groups, policy actions can be designed to be more specific and consequently more effective. Therefore, in this study, we attempt first to determine the extent of the preventable hospitalization problem in Harris County, followed by examining the disparities in these hospitalizations by sociodemographic characteristics and geographic locations. Finally, we attempt to determine how improving access to primary care can be expected to reduce the individual risk and ZIP code level rates of these hospitalizations. 3

4 SPECIFIC OBJECTIVES Objective 1: Observe trends in the overall and disease-specific rates of preventable hospitalizations over time. Objective 2: Explore the variation in and persistence of rates of preventable hospitalizations by ZIP-code and by individual characteristics. Objective 3: Observe changes in the rates of multiple admissions for preventable hospitalizations over time. Objective 4: Examine the unadjusted effect of primary care safety net and private primary care physician availability on risk and rate of preventable hospitalizations in the local safety net population. Objective 5: Determine the effect of primary care safety net and private primary care physician availability on risk and rate of preventable hospitalizations in the local safety net population, while adjusting for individual and ZIP-code level characteristics. 4

5 METHODS Study Design This study was a retrospective cross-sectional analysis of a state-level hospital inpatient discharge dataset. Study Population This study was restricted to residents of Harris County, Texas, age 18 to 64, who had at least one hospitalization in the years 2003 to Access to primary care is mainly an issue in the uninsured and the publicly insured. Therefore, in studying the effect of primary care safety net and physician availability on PHs, the population of interest was the uninsured and the Medicaid-insured. This population is also known as the safety net population. Datasets 1. Texas Health Care Information Collection (THCIC) Hospital Discharge data Information on preventable hospitalizations and personal social and demographic characteristics of hospitalized individuals was obtained from the THCIC inpatient discharge data. The THCIC data provides information on all hospital discharges from all state-licensed hospitals in Texas, barring certain exceptions. The THCIC data are available in a Public Use Data File (PUDF) and a Research Use Data File (RUDF). The PUDF contains information on demographic characteristics, geographic location, diagnosis, procedures, and sources of payment for individuals who have had an inpatient discharge from a Texas hospital. Some information that is collected but suppressed or unavailable in the PUDF can be obtained from the Research Files. For the purpose of this study, THCIC data for the years 2003 to 2008 was used. 2. Census 2000 Census 2000 was used to obtain population estimates for calculating PH rates among non-elderly adults in Harris County for year 2003 to ZIP code level estimates for household income in year 2000 were also obtained from the Census. 3. St.Luke s Episcopal Health Charities Project Safety Net Data 2009 The Project Safety Net (PSN) survey is a community-based program, managed by the St.Luke s Episcopal Health Charities. This survey was first conducted in 2004, and then again in The 2009 survey was designed to collect information on the basic clinic characteristics as well as primary care capacity for all the public and private safety net providers in Harris County in 2007, 2008 and In this study, the PSN data was used to determine the location of primary care safety net clinics in Harris County to measure the availability of primary care safety net clinics. 4. Texas Medical Board (TMB) Complete Electronic Database 2010 The TMB electronic data is a complete database of all healthcare practitioners, including physicians, with an active Texas license, residing both in-state and out-of-state. The information in the dataset includes name of practitioner, license number, mailing address, 5

6 office address, birth year and place, gender, race and several other characteristics. For physicians, the data includes additional information on degree, specialty (primary and secondary), practice type, practice setting, etc. In this study, this database will be used to obtain names and office addresses of all primary care physicians practicing in Harris County. Dependent Variable Preventable Hospitalizations The primary dependent variable for this study was Preventable Hospitalizations (PHs). It was defined as a binary variable depending on whether a given hospitalization was preventable ( 1 ) or not ( 0 ). PHs were identified based on the Agency of Healthcare Research and Quality classification system (AHRQ, 2007). All hospitalizations that were categorized as a transfer from another hospital were excluded to prevent double-counting. Also, since the study population excluded children, all hospitalizations for Low Birth Weight (which is one of the AHRQ classified PHs) were excluded. Consequently a total of 13 types of hospitalizations were identified as PHs. In identifying PHs, first a binary variable for each of the 13 types of PHs was created using the AHRQ classification system. For a specific type of PH, the corresponding binary variable was coded as 1, if the ICD-9-CM diagnosis code for the hospitalization met the AHRQ definition. Otherwise, the variable was coded as zero. Finally, a binary variable was created for overall PH, coded as 1 if at least one of the disease-specific PHs is coded as 1, and zero otherwise. Independent Variables Primary care safety net availability and private primary care physician availability were the primary independent variables in the study. Primary Care Safety Net Availability For the purpose of this study, a primary care safety net clinic was defined as a clinic included in the Project Safety Net survey that provides primary care services to Harris County residents, serves the uninsured, serves non-elderly adults, was functional in the year 2008 and is open at least 20 hours a week. A list of eligible primary care safety net clinics was obtained from the Project Safety Net data along with their addresses. Availability of primary care safety nets was measured as the number of safety net clinics located within a 5-mile radius of the centroid of the patient s ZIP code of residence per 1000 population. Latitude and longitude coordinates for the safety net clinics and the centroids were obtained with the help of Pat Courtney using the ArcGIS software at the School of Public Health. The Great Circle Distance formula was used to calculate the distance between a safety net clinic and the ZIP code centroid. According to this formula Distance = * arccos[sin(lat1/ ) * sin(lat2/ ) + cos(lat1/ ) * cos(lat2/ ) * cos(lon2/ lon1/ )] 6

7 For each ZIP code, the total number of safety net clinics within a 5-mile was enumerated, which was then divided by the Census 2000 estimates of the ZIP code population between ages 18 and 64 to obtain primary care safety net availability per 1000 population in the ZIP code. Primary Care Physician Availability A private primary care physician was defined as a general physician, internist or family practitioner, with a license to practice in the state of Texas in 2008 and an active practice in either Harris or one of the adjacent counties, who is involved in patient care at least 50% of the time ( 20 hours per week) and does not have the same office address as a safety net clinic. A list of eligible primary care physicians was obtained from the Texas Medical Board data. Availability of primary care physicians was measured as the number of primary care physicians located within a 5-mile radius of the centroid of the patient s ZIP code of residence per 1000 population. Latitude and longitude coordinates for the physician s office and the centroid were obtained with the help of Pat Courtney using the ArcGIS software at the School of Public Health. The Great Circle Distance formula was used to calculate the distance between a physician s office address and the ZIP code centroid. For each ZIP code, the total number of physicians within a 5-mile was enumerated, which was then divided by the Census 2000 estimates of the ZIP code population between ages 18 and 64 to obtain primary care physician availability per 1000 population in the ZIP code. Other independent variables measured in the study were sociodemographic characteristics like age, gender, race, ethnicity, insurance status and median household income of the ZIP code of residence. In determining insurance status, the primary payment source listed in the PUDF was used. All hospitalizations where the primary payment source was listed as selfpay or charity/indigent were classified as Uninsured. If the primary payment source was Medicaid, the hospitalization was categorized as Medicaid-insured. All other hospitalizations were considered to be Privately Insured. All dependent and independent variables used in the study are listed in Table 2. 7

8 Table 2 Study Variables Measured Variable Source Definition Preventable Hospitalizations Primary Care Safety Net Availability Primary Care Physician Availability THCIC Dependent Variable Independent Variables THCIC Project Safety Net THCIC TMB Complete Electronic Dataset Any hospitalization that meets at least one of the 13 AHRQ criteria for PHs Binary Variable 1 Preventable Hospitalization 0 Non-preventable Hospitalization Number of primary care safety net clinics within a 5-mile radius of a patient s ZIP code of residence Number of primary care physicians within a 5-mile radius of a patient s ZIP code of residence Age THCIC Age of patient at the time of hospital discharge Categorical Variable 0 18 to to to 64 Gender THCIC Gender of patient Binary Variable 0 Male 1 Female Race THCIC Race of patient Categorical Variable 0 Non-White 1 White Ethnicity THCIC 0 Non-Hispanic 1 Hispanic Health Insurance Status THCIC Health insurance status of patient at the time of discharge from hospital. Binary Variable 0 Privately Insured 1 Uninsured 2 Medicaid-insured Income THCIC Census 2000 SF-3 Median income in the patient s ZIP code of residence. 8

9 Data Analysis Trends in the overall and disease-specific rates of preventable hospitalizations The binary PH variables were used to determine the total number of PHs among nonelderly adults in Harris County for the years 2003 to Both overall and disease-specific PHs were enumerated. The total number of all hospitalizations (PHs and non-phs) was also determined for each year. Rate of PHs (overall and disease-specific) was calculated in two ways 1. Number of PHs per 100,000 non-elderly adult population of Harris County Rate of PHs = (No of PHs/Total non-elderly adult population)*100, Number of PHs per 100 hospitalizations Rate of PHs = (No of PHs/Total number of hospitalizations)*100 Rates of PHs were calculated using two different methods because the first method would mask any change in PHs that actually occurs due to changes in overall hospitalization rates. For example, if PHs rates per 100,000 are increasing, it may be due to a true increase in PH risk among the population, or due to simply an overall increase in the tendency to be hospitalized. The rate of PHs per 100 hospitalizations allows us to adjust for this confounding phenomenon. Persistence of Preventable Hospitalizations by Geographic Location (ZIP codes) and Sociodemographic Characteristics The number of PHs in each ZIP code in Harris County was determined, and PH rates per 100,000 population and per 100 hospitalizations were calculated. The ZIP codes were then categorized into quintiles based on their PH rates in 2003, with the first quintile including ZIP codes with lowest PH rates, and the last quintile including ZIP codes with the highest PH rates, and persistence of PH rates within these quintiles over the years was observed. Univariable analysis of variance (ANOVA) was used to examine if the PH rates differed significantly between quintiles in a given year and within a quintile over 6 years. ZIP codes 77010, and were excluded as they had very low population numbers and consequently highly imprecise rates. Variability in PH rates by age, gender, race, ethnicity and insurance status over the years was also examined. The number of PHs in each sociodemographic group was obtained for each year from 2003 to 2008 and overall PH rates per 100,000 population and per 100 hospitalizations were calculated. The only exception was made when comparing PH rates by insurance status. Since county-level population counts for each insurance group were not available, PH rates were only measured per 100 hospitalizations for comparison by insurance status. 9

10 Trends in the rates of multiple preventable hospitalizations Multiple preventable hospitalizations (MPHs) were observed when the same individual had more than one PH in the same year. In defining MPHs, the first PH for an individual in a given year was considered as the index admission. All PHs after the index PH in the given year were considered as MPHs. Certain exclusion criteria were used while identifying MPHs - 1. If the next hospitalization occurred on the same day as the previous one, it was considered as a transfer and not a multiple hospitalization. 2. Hospitalizations where the date of admission was missing. 3. Hospitalizations where the patient was admitted before November 1 of the previous year. 4. Hospitalizations where patients with the same identification number had different age, gender and/or addresses. The total number of MPHs (overall and for each specific ACSC) was determined and the rate of MPHs was calculated as the number of MPHs per 100 PHs. The total number of patients with MPHs and their percentage in the total population with PHs was also determined. Trends in the rates of MPHs as well as in the percentage of patients with MPHs were examined from 2003 to Disparities in MPH rates by age, gender, race, ethnicity and insurance status were also examined. Effect of primary care safety net and primary care physician availability on the risk and rate of PHs in the local safety net population Safety net clinics generally serve the primary care needs of the low-income individuals in the community, specifically the uninsured and the publicly insured. Therefore, the local safety net population was defined as non-elderly individuals who are uninsured or Medicaid-insured. The primary independent variables for this analysis were primary care safety net availability and primary care physician availability, which were defined as mentioned earlier. Risk of a PH was measured at the discharge level and was defined as the probability that a given hospitalization is a PH. Rate of a PH was measured at the ZIP code level, and was measured as number of PHs per 100 hospitalizations in a ZIP code. A univariable logistic regression analysis was used to determine the unadjusted effect of primary care safety net and of primary care physician availability on the individual risk of PHs. A univariable linear regression analysis was used to determine the unadjusted effect of these factors on the rate of PHs in a given ZIP code. 10

11 Effect of primary care safety net and primary care physician availability on the risk and rate of PHs in the local safety net population, adjusted for individual and community level characteristics. Multivariable logistic regression analysis was used to determine the effect of primary care safety net and physician availability on the risk of PHs, while controlling for age, gender, race, ethnicity, insurance status and income. Three different multivariable logistic regression models were constructed expressing PH risk as a product of the following factors Model 1 Safety net availability + sociodemographic factors Model 2 Physician availability + sociodemographic factors Model 3 Safety net availability + Physician availability + sociodemographic factors Multivariable linear regression analysis was used to determine the effect of primary care safety net and physician availability on the rate of PHs, while controlling for age, gender, race, ethnicity, and income at the ZIP code level. Three different multivariable linear regression models were constructed, expressing PH rate as a product of the following factors Model 1 Safety net availability + sociodemographic factors Model 2 Physician availability + sociodemographic factors Model 3 Safety net availability + Physician availability + sociodemographic factors 11

12 RESULTS AND CONCLUSIONS Trends in the overall and disease-specific rates of preventable hospitalizations Between 2003 and 2008, the number and rate of PHs per 100,000 population in Harris County have consistently decreased (Table 3, Figure 1 and 2). The number fell from 18,651 PHs in 2003 to 16,798 in The rate of PHs per 100,000 population decreased from about 810 hospitalizations per 100,000 population to 667 hospitalizations per 100,000 population. But this could be misleading as the number and rate of hospitalizations for all causes in Harris County have also declined over the years (Table 3). An examination of rate of PHs per 100 hospitalizations showed a very slight decline (Table 4; Figure 3). Therefore, most of the decline in PHs over the years is mainly due to a decline in overall hospitalization rates. When PHs for each specific ACSC were examined, most conditions showed a decline in the number and rate of hospitalizations (Table 3). Over the years, congestive heart failure (CHF) has consistently been the most common type of PH (Table 4). In 2008, there were hospitalizations for CHF per 100,000 population and they made up 20.19% of all PHs. Bacterial pneumonia and long-term complications for diabetes are other two conditions that have had consistently high rates of hospitalization. 12

13 Table 3 Rates of Preventable Hospitalizations in Harris County in Adults 18 to 64 ( ) 2003 a 2004 a Hospitalizations No. of Hosp. PH Rate b No. of Hosp. PH Rate b No. of Hosp. All Hospitalizations 230, , , , , , All Preventable Hosp. 18, , , , , , Diabetes Short-term Com 1, , , , , , Perforated Appendix c Diabetes Long-term Com. PH Rate b No. of Hosp. PH Rate b No. of Hosp. PH Rate b No. of Hosp. 1, , , , , , Chronic Obst. Pulm. Dis. 1, , , , , , Hypertension 1, , , , , Cong. Heart Failure 3, , , , , , Dehydration 1, Bacterial Pneumonia 3, , , , , , Urinary Tract Infection 1, , , , , , Angina w/o Procedure Uncontrolled Diabetes Adult Asthma 1, , , , , , Lower Extremity Amputation in Diabetes a Number of preventable hospitalizations adjusted for smaller number of diagnosis categories b PH rate calculated per 100,000 population c Rate calculated per 100 hospital admissions for appendicitis (according to AHRQ definition) PH Rate b 13

14 Table 4 Proportion of Preventable Hospitalizations in Harris County in Adults 18 to 64 ( ) 2003 a 2004 a Hospitalizations No. of Hosp. Percent No. of Hosp. Percent No. of Hosp. Percent No. of Hosp. Percent No. of Hosp. Percent All Hospitalizations 230, , , , , , All Preventable Hosp. b 18, , , , , , Diabetes Short-term Com 1, , , , , , Perforated Appendix c Diabetes Long-term Com. No. of Hosp. 1, , , , , , Chronic Obst. Pulm. Dis. 1, , , , , , Hypertension 1, , , , , Cong. Heart Failure 3, , , , , , Dehydration 1, Bacterial Pneumonia 3, , , , , , Urinary Tract Infection 1, , , , , , Angina w/o Procedure Uncontrolled Diabetes Adult Asthma 1, , , , , , Lower Extremity Amputation in Diabetes a Number of preventable hospitalizations adjusted for smaller number of diagnosis categories b For overall PHs, PH percentage is equivalent to PH rate per 100 hospitalizations Percent 14

15 Preventable Hospitalization per 100,000 Number of Preventable Hospitalizations Figure 1: Number of Preventable Hospitalizations in Harris County ( ) 20,000 19,000 18,000 17,000 16,000 15,000 Figure 2: Trends in Preventable Hospitalization Rates per 100,000 population ( )

16 Preventable Hospitalizations per 100 Hospitalizations Figure 3: Trends in Preventable Hospitalization Rates per 100 Hospitalizations ( )

17 Persistence of Preventable Hospitalizations by Geographic Location (ZIP codes) and Sociodemographic Characteristics Persistence of Preventable Hospitalizations by Geographic Location (ZIP codes) The PH rates per 100,000 population were visibly as well as statistically significantly different between the quintiles in 2003, and remained so for the most part over the years (Table 5, Figure 4). The PH rates for quintiles 1 to 4 remained relatively constant, with slight declines from 2003 to On the other hand, the PH rates in the fifth quintile (highest PH rates) declined continuously over the years from an average rate 1499 per 100,000 in 2003 to 1169 per 100,000 in 2008, although this decline was not statistically significant. Despite this decline, the PH rates in the fifth quintile remained significantly higher than the PH rates in all the other quintiles. Similar patterns were seen when the rate of PH per 100 hospitalizations was observed (Table 6, Figure 5). A visual representation of the distribution of PHs among various ZIP codes in Harris County is given in the form of a map (Figure 6). The map indicates that PHs are not equally distributed through all the ZIP codes of Harris County and that ZIP codes with low or high PH rates tend to cluster together, highlighting broad geographical regions of health care access shortage. The highest rates of PHs are seen in the eastern region of the county, especially the north-eastern borders. Table 5: Preventable Hospitalization Rates per 100,000 population by ZIP code Quintile Quintile Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile a ab abc abcd a ab abc abcd ab ab abcd ab ab abcd ab ab abcd ab ab abcd a Significant difference compared to Quintile 1; b Significant difference compared to Quintile 2 c Significant difference compared to Quintile 3; d Significant difference compared to Quintile 4 17

18 Hospitalizations per 100,000 Table 6: Preventable Hospitalization Rates per 100 hospitalization by ZIP code Quintile Quintile Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile a ab abc abcd a a abc abcd a a abc abcd a a abc abcd a a abc abcd a a abc abcd a Significant difference compared to Quintile 1; b Significant difference compared to Quintile 2 c Significant difference compared to Quintile 3; d Significant difference compared to Quintile 4 Figure 4: Trends in ZIP-code Level PH Rates per 100,000 population Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile

19 Proportion of Hospitalizations Figure 5: Trends in ZIP-code Level PH Rates per 100 Hospitalizations Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile Figure 6 Geographical Distribution of PHs among Harris County ZIP codes Source Texas Health Care Information Collection

20 Preventable Hospitalizations by Demographic Characteristics Age: The rates of PHs measured per 100,000 population as well as per 100 hospitalizations were lowest in the youngest age group (18-34) and increased constantly with increasing age (Table 7, Figure 7). Also, there is a steep decline in PH rates per 100,000 population in the oldest age group over the years, while the rates in the younger age groups remain relatively constant. Table 7: Age Distribution in Preventable Hospitalizations Hospitalizations among individuals aged No. of Hosp. No. of PHs Rate of Hosp. (per 100,000 pop.) Rate of PHs (per 100,000 pop.) Rate of PHs (per 100 Hosp.) Hospitalizations among individuals aged No. of Hosp. No. of PHs Rate of Hosp. (per 100,000 pop.) Rate of PHs (per 100,000 pop.) Rate of PHs (per 100 Hosp.) Hospitalizations among individuals aged No. of Hosp. No. of PHs Rate of Hosp. (per 100,000 pop.) Rate of PHs (per 100,000 pop.) Rate of PHs (per 100 Hosp.)

21 Preventable Hospitalizations per 100 Hospitalizations Preventable Hospitalizations per 100,000 population Figure 7: Trends in PH rates (per 100,000 population) by Age to to to 64 Figure 8: Trends in PH Rates (per 100 Hospitalizations) by Age to to to

22 Gender: Overall, females had a higher rate of hospitalizations and of preventable hospitalizations when measured per 100,000 population. But, when PH rate was measured per 100 hospitalizations, males had higher rates as compared to females, indicating that although females had a higher tendency to be hospitalized, males had a higher risk of having a PH. Table 8: Gender Distribution in Preventable Hospitalizations Hospitalizations among Males No. of Hosp. No. of PHs Rate of Hosp. (per 100,000 pop.) Rate of PHs (per 100,000 pop.) Hospitalizations among Females No. of Hosp. No. of PHs Rate of Hosp (per 100,000 pop.) Rate of PHs (per 100 Hosp.) Rate of PHs (per 100,000 pop.) Rate of PHs (per 100 Hosp.) 22

23 Preventable Hospitalizations per 100 Hospitalizations Preventable Hospitalizations per 100,000 population Figure 9: Trends in PH Rates (per 100,000 population) by Gender Female Male Figure 10: Trends in PH Rates (per 100 hospitalizations) by Gender Female Male

24 Race: White race was consistently associated with low PH rates. In fact, when PH rates were measured per 100,000 population, non-whites had almost 4 times higher PH rates than whites (Table 9, Figure 11). This disparity was smaller when the rate was measured per 100 hospitalizations but was still extremely prominent (Table 9, Figure 12). Table 9: Racial distribution in Preventable Hospitalizations Hospitalizations among Whites No. of Hosp. No. of PHs Rate of Hosp (per 100,000 pop.) Rate of PHs (per 100,000 pop.) Hospitalizations among Non-Whites No. of Hosp. No. of PHs Rate of Hosp (per 100,000 pop.) Rate of PHs (per 100 Hosp.) Rate of PHs (per 100,000 pop.) Rate of PHs (per 100 Hosp.) 24

25 Preventable Hospitalizations per 100 Hospitalizations Preventable Hospitalizations per 100,000 population Figure 11: Trends in PH Rates (per 100,000 population) by Race White Non-white Figure 12: Trends in PH Rates (per 100 hospitalizations) by Race White Non-white

26 Ethnicity: Hispanics had lower rates of preventable hospitalizations as compared to their non- Hispanic counterparts. PH rates in non-hispanics were about twice of those in Hispanics when measured per 100,000 population (Table 10, Figure 13). The disparity in PH rates was less pronounced when measured per 100 hospitalizations, where rates in non-hispanics were about 50% higher (Table 10, Figure 14). Table 10: Ethnic Distribution in Preventable Hospitalizations Hospitalizations among Hispanics No. of Hosp. No. of PHs Rate of Hosp (per 100,000 pop.) Rate of PHs (per 100,000 pop.) Rate of PHs (per 100 Hosp.) Hospitalizations among Non-Hispanics No. of Hosp. No. of PHs Rate of Hosp (per 100,000 pop.) Rate of PHs (per 100,000 pop.) Rate of PHs (per 100 Hosp.)

27 Preventable Hospitalizations per 100 Hospitalizations Preventable Hospitalizations per 100,000 population Figure 13: Trends in PH Rates (per 100,000 population) by Ethnicity Hispanic Non-Hispanic Figure 14: Trends in PH Rates (per 100 hospitalizations) by Ethnicity Hispanic Non-Hispanic

28 Insurance Status: The uninsured had the highest rate of preventable hospitalizations with more than 10% of their hospitalizations (10.84 PHs per 100 hospitalizations) being preventable in nature in On the other hand, the Medicaid-insured had the lowest rate, with only about 6.25% preventable hospitalizations, which further reduced to 4.5% in All other insured individuals lay somewhere in the middle with 8% of their hospitalizations categorized as preventable. These differences do not change much over time, indicating that the uninsured in Harris County face a persistent lack of primary care access, despite the presence of a safety net system. Table 11: Preventable Hospitalizations by Insurance Status Hospitalizations among the Uninsured No. of Hosp No. of Prev. Hosp PH Rate (per 100 Hosp.) Hospitalizations among the Medicaid-Insured No. of Hosp No. of Prev. Hosp PH Rate (per 100 Hosp.) Hospitalizations among all Other Insured No. of Hosp No. of Prev. Hosp PH Rate (per 100 Hosp.)

29 Preventable Hospitalizations per 100 Hospitalizations Figure 15: Trends in PH Rates (per 100 Hospitalizations) by Insurance Status Uninsured Medicaid Other Insured 2 0 When hospitalization rates for individual ACSCs were examined by insurance status, hospitalizations for most ACSCs showed trends similar to the overall trend. But a few conditions showed patterns that indicated either the presence of larger or smaller disparities by insurance status, as compared to the overall trend. Although overall trends showed that the proportion of PHs among all hospitalizations in the uninsured was about 1.75 times higher than the Medicaid-insured and about 1.25 times higher than other insured, the proportions for shortterm complications of diabetes and perforated appendicitis were almost 3 times higher among the uninsured as compared to the other two groups (Figure 16 and 17). Thus, disparities among the insured and the uninsured were larger for these two conditions, marking them as significant problem areas for access. On the other hand, the uninsured, Medicaid-insured and other insured had almost equivalent proportions of hospitalizations for chronic obstructive pulmonary disorder, congestive heart failure and bacterial pneumonia, indicating no disparities by insurance status in hospitalization for these conditions (Figures 18 20). The insured had a slightly higher proportion of hospitalizations for dehydration as compared to the uninsured and the Medicaidinsured (Figure 21) 29

30 Preventable Hospitalizations per 100 Hospitalizations Preventable Hospitalizations per 100 Hospitalizations Figure 16: Hospitalizations for Short-term Complications of Diabetes (per 100 Hospitalizations) by Insurance Status Uninsured Medicaid Other Insured Figure 17: Hospitalizations for Perforated Appendicitis as a Percentage of All Hospitalizations examined by Insurance Status Uninsured Medicaid Other Insured

31 Preventable Hospitalization per 100 Hospitalizations Preventable Hospitalizations per 100 Hospitalizations Figure 18: Hospitalizations for Chronic Obstructive Pulmonary Disorder (per 100 Hospitalizations) by Insurance Status Uninsured Medicaid Other Insured Figure 19: Hospitalizations for Congestive Heart Failure (per 100 Hospitalizations) by Insurance Status Uninsured Medicaid Other Insured

32 Preventable Hospitalizations per 100 Hospitalizations Preventable Hospitalizations per 100 Hospitalizations Figure 20: Hospitalizations for Bacterial Pneumonia (per 100 Hospitalizations) by Insurance Status Uninsured Medicaid Other Insured Figure 21: Hospitalizations for Dehydration (per 100 Hospitalizations) by Insurance Status Uninsured Medicaid Other Insured

33 Number of Multiple PHs per 100 PHs Trends in the Rates of Multiple Preventable Hospitalizations In 2003, about 3700 PHs were repeat hospitalizations, giving a multiple PH rate of about 20 PHs per 100 PHs i.e. about 20% of all PHs are repeat hospitalizations. When trends over the years were examined, these rates remained unchanged, indicating a persistent multiple PH problem (Table 12, Figure 22). All these multiple PHs were caused by a set of 2200 people in These patients constituted about 15% of all the patients who were admitted with a PH i.e. one in every six PH patients is admitted for more than one PH in a given year. These numbers remain unchanged over the years (Table 12, Figure 23). Hence, these patients may be facing a persistent lack of access to primary care which does not resolve even after having one PH. Table 12: Multiple Preventable Hospitalizations in non-elderly adults in Harris County Multiple Preventable Hosp. Patients with Multiple Prev. Hosp. Avg. No. of No. of MPHs Total No. of PHs Rate of MPHs No. of Patients with MPHs Total No. of Patients Hosp. per patient Percent patients with MPHs with PHs , , , , , , Figure 22: Trends in Rate of Multiple Preventable Hospitalizations

34 Patients with Multiple PHs per 100 PH patients Figure 23: Trends in Individual Patients with Multiple Preventable Hospitalizations Multiple Preventable Hospitalizations by Disease Condition When multiple PHs were examined by disease condition, significant variability in rates was observed (Table 13). Conditions like hypertension, dehydration, bacterial pneumonia and UTIs had very low rates of multiple hospitalizations (between 4 and 7 percent). On the other hand, conditions like COPD and CHF had multiple hospitalization rates that were higher than the overall rates. CHF had the highest multiple PH rate of about 25 PHs per 100 PHs, which stayed consistent through the years. CHF was also the condition that contributed the largest proportion of all multiple hospitalizations (almost 25% of all multiple hospitalization were for CHF data not shown). Multiple Preventable Hospitalizations by Individual Characteristics Significant disparities in multiple PHs were observed by sociodemographic characteristics (Table 14). The rate of multiple PHs increased with age, with more than 17% of patients in the oldest age group returning to the hospital for preventable admissions in 2008 (Figure 24). Multiple PH rates did not differ visibly by gender (Figure 25). Non-whites had higher rates of multiple PHs than whites, whereas non-hispanics were more prone to multiple PHs than Hispanics (Figures 26, 27). Examination of disparities by insurance status also revealed significant disparities, with the Medicaid-insured reporting almost two times higher rates of multiple PHs as compared to the uninsured and the privately insured (Figure 28). 34

35 Table 13: Multiple Preventable Hospitalizations (Disease Specific) among non-elderly adults in Harris County Disease Condition a,b No. of Rate c No. of Rate No. of Rate No. of Rate No. of Rate No. of Rate MPHs MPHs MPHs MPHs MPHs MPHs Diabetes ST Com Diabetes LT Com COPD Hypertension Cong. Heart Failure Dehydration Bacterial Pneumonia UTI Adult Asthma LEA in Diabetes a Does not include admissions for perforated appendicitis as multiple admissions for this condition are not plausible. b Does not include admissions for Angina without Procedure and Uncontrolled Diabetes because of very low numbers. c Rate is measured as number of Multiple PHs (MPHs) per 100 PHs 35

36 Table 14: Multiple Preventable Hospitalizations by Individual Sociodemographic Characteristics Patient Characteristic Age 18 to to to 64 Gender Male Female Race White Non-White Ethnicity Hispanic Non-Hispanic Health Insurance Uninsured Medicaid-Insured Other Insured No. of MPHs Rate a No. of MPHs Rate a a Rate is measured as number of Multiple PHs (MPHs) per 100 PHs No. of MPHs Rate a No. of MPHs Rate a No. of MPHs Rate a No. of MPHs Rate a

37 MPHs per 100 PHs MPHs per 100 PHs Figure 24: Trends in Multiple Preventable Hospitalizations by Age s s s 0 Figure 25: Trends in Multiple Preventable Hospitalizations by Gender Males Females

38 MPHs per 100 PHs MPHs per 100 PHs Figure 26: Trends in Multiple Preventable Hospitalizations by Race Non-Whites Whites 5 0 Figure 27: Trends in Multiple Preventable Hospitalizations by Ethnicity Non-Hispanic Hispanic

39 MPHs per 100 PHs Figure 28: Trends in Multiple Preventable Hospitalizations by Insurance Status Insured Uninsured Medicaid

40 Effect of Primary Care Safety Net and Primary Care Physician Availability on Risk and Rate of PHs in the Local Safety Net Population Profile of Primary Care Safety Net Clinics: The PSN survey included a total of 133 safety net clinics, of which only 40 were eligible for inclusion in our study. These clinics included federally qualified health clinics (FQHCs), Harris County Hospital District clinics and private not-for-profit clinics. The profile of all clinics included in the study is outlined in Table 15. FQHCs were the most common type of clinic, whereas HCHD clinics tended to be the largest in size. Table 15: Profile of Safety Net Clinics Clinic type Frequency (%) Avg. Clinic Size (sq ft) FQHC 20 (50.00%) Hospital District 12 (30.00%) Private Not-for-profit 8 (20.00%) Total 40 (100%) Profile of Primary Care Physicians: There were a total of 2665 primary care physicians in Harris and neighboring counties who provided primary care services in the year Of these, more than 1645 physicians practiced in Houston. Most physicians were internal medicine practitioners (53.66%), followed by family medicine/practice (41.01%) and general physicians (5.33%). Preventable Hospitalizations Among Local Non-Elderly Adult Safety Net Population: There were a total of hospitalizations among the non-elderly adult safety net population of Harris County in the year 2008 (Table 16), of which 6315 were PHs. Thus, on an average, there was a 7% probability that a hospitalization in a non-elderly adult resident of Harris County from the safety net population could be a PH. When PHs were examined at the ZIP code level, all ZIP codes (except one) had at least one PH in the year The average rate of PHs was 7.12 PHs per 100 hospitalizations in a ZIP code. ZIP code had the lowest PH rate of zero PHs per 100 hospitalizations, whereas ZIP code had the highest PH rate of 13.5 PHs per 100 hospitalizations. ZIP codes 77010, and had very low population counts and consequently provided highly imprecise estimates of PHs. Therefore, all hospitalizations from these ZIP codes were excluded from further analysis. A total of 137 ZIP codes were included in the final analysis. 40

41 Effect of Safety Net Availability and Physician Availability Univariable Analysis: First, the effect of safety net availability and primary care physician availability on the risk and rate of PHs were examined in isolation in univariable logistic and linear regression models. Risk of Preventable Hospitalizations On an average, there were about 0.2 safety net clinics and 9.3 physicians available to residents of Harris County per 1000 population (Table 16). In the unadjusted analysis, higher availability of safety net clinics and primary care physicians actually increased the risk of PHs. A unit increase in the number of safety nets per 1000 population increased the odds of a PH by about 80%. A unit increase in the number of physicians per 1000 population increased the odds of a PH by 0.2% (Table 16). Table 16: Individual Characteristics and Risk of Preventable Hospitalizations Preventable Hospitalizations Non-Preventable Hospitalizations All Hospitalizations Risk of PHs Total Age Gender Male Female Race Non-White White Ethnicity Non-Hisp. Hispanic Insurance Medicaid Uninsured Income (in thousands) Mean (SD) Safety Net Availability (per 1000 population) Physician Availability (per 1000 population * p< (13.168) (0.359) (17.938) (13.914) (0.334) (16.498) (13.865) (0.336) (16.608) Unadj. Odds Ratio Reference 4.420* 8.504* Reference 0.336* Reference 0.736* Reference 0.459* Reference 2.530* * * * 41

42 The effects of age, gender, race, ethnicity, insurance status and income on the risk of PHs were also examined in individual univariable logistic regression models. There were significant differences in an individual s risk of PHs based on their socio-demographic group. Table 16 summarizes the unadjusted differences in risk of PH by individual characteristics. Uninsured individuals were at a higher risk of PHs than the Medicaid-insured. Also, males, non-whites and non-hispanics were at higher risk of PHs than females, whites and Hispanics respectively. Lower ZIP code-level income was also significantly associated with higher PH probability. Rate of Preventable Hospitalizations On an average, each ZIP code had about 0.24 safety net clinics and physicians available per 1000 population. A univariable analysis of effect of safety net availability and primary care physician availability on ZIP code level rates of PHs failed to reveal any significant differences (Table 17). The effects of age, gender, race, ethnicity and income distributions of a ZIP code on the corresponding rate of PHs were also examined in individual univariable linear regression models. The unadjusted effects of ZIP code sociodemographic characteristics on PH rates are summarized in Table 17. Only the percentage of whites as well as the median income of a ZIP code had statistically significant effects on PH rates, with higher levels of both being associated with lower PH rates in a ZIP code. Table 17: ZIP Code Characteristics and Preventable Hospitalization Rate Variable Unadj. Parameter Standard Error t-value p-value Estimate Median Age Female Percent White Percent < Hispanic Percent Median Income (in < thousands) Safety Net Availability (per 1000 population) Physician Availability (per 1000 population)

43 Effect of Safety Net Availability and Physician Availability Multivariable Analysis Risk of Preventable Hospitalizations Three different models were constructed to examine the effect of primary care availability on the risk of PHs, while adjusting for sociodemographic characteristics of the individual. All three models showed a statistically significant negative association between primary care availability and the risk of PHs (Table 18). In model 1, an increase in safety net clinics available to an individual by 1 per 1000 population was associated with a 25% reduction in the odds of a PH. In model 2, an increase in physician availability by 10 per 1000 population was expected to lead to a 6% reduction in the odds of a PH. The associations remained significant even when the effect of safety net and physician availability were examined together (Model 3), although their strength reduced a little. A unit increase in number of safety nets per 1000 population was associated with a 14% reduction in odds of a PH, whereas an increase in physician availability by 10 per 1000 population led to a 4% reduction in odds. Table 18: Multivariable Logistic Regression for Effect of Primary Care Availability on Risk of Preventable Hospitalizations Variable Model 1 Model 2 Model 3 Safety Net Availability (per ( ) ( ) 1000 population) Physician Availability (per ( ) ( ) 1000 population) Age ( ) ( ) ( ) Age ( ) ( ) ( ) Female ( ) ( ) ( ) White ( ) ( ) ( ) Hispanic ( ) ( ) ( ) Uninsurance ( ) ( ) ( ) Income (in thousands) ( ) ( ) ( ) Rate of Preventable Hospitalizations Three different models were constructed to examine the effect of primary care availability on the ZIP code rates of PHs, while adjusting for sociodemographic characteristics of the ZIP code. In all three models, primary care availability, measured as safety net and physician availability was not significantly associated with the rate of PHs in a ZIP code (Table 19). The only factors that were found to have a statistically significant effect on ZIP code rates of PHs were the proportion of Hispanics in a ZIP code as well as the median household income of the ZIP code. 43

44 Table 19: Multivariable Linear Regression for Effect of Primary Care Availability on Rate of Preventable Hospitalizations in a ZIP code Variable Model 1 Model 2 Model 3 Safety Net Availability (per 1000 population) (0.5186) (0.7484) Physician Availability (per 1000 population) (0.0081) (0.0117) Median Age (0.0832) (0.0809) (0.0829) Female Percent (0.0619) (0.0595) (0.0616) White Percent (0.0110) (0.0108)* (0.0109) Hispanic Percent (0.0130)* (0.0124)* (0.0134)* Median Income (in thousands) (0.0157)* (0.0154)* (0.0156)* Geographical Correlation between PH rates and Locations of Safety Nets and Physicians: A visual inspection of actual locations of safety nets and physicians relative to ZIP code level PH rates (per 100 hospitalizations) was also conducted by construction ZIP code level maps. The maps reveal that primary care safety nets and physicians tend to be concentrated near the center of the city, especially in the Medical Center and the Downtown area (Figures 29 and 30). Although these areas do contain some ZIP codes with high PH rates, most of the high PH ZIP codes are located near the outskirts of the city, especially on the northeastern side (Figure 29). But this area has a very low supply of safety nets as well as physicians. Therefore, efforts should be made to expand primary care resources in a manner that the problem areas can be provided with good primary care access. 44

45 Figure 29: Preventable Hospitalization Rates and Location of Safety Net Clinics Source Texas Health Care Information Collection ; Project Safety Net Survey

46 Figure 30: Locations of Private Primary Care Physicians by ZIP code Source Texas Medical Board Complete Electronic Dataset