Purpose of Report
The purpose of the State of the State’s Health Report is to provide readers with information regarding the health status of Oklahoma residents. The report presents data on overall deaths, infant deaths, and leading causes of death; rates of some chronic diseases; and rates of several health behaviors and risk factors for chronic diseases. Grades are assigned to data for each demographic and geographic group to enable readers to view patterns that occur for each indicator. Differences between groups are not statistically tested, and assumptions regarding statistically significant differences should not be made.
Selection of Health Indicators
Health indicators for the State of the State’s Health Report were chosen based on practical considerations regarding certain qualities of the indicators. In general terms, health indicators were selected for the report when one or more of the following conditions were evident:
- There was a perceived ability to effect change in the health indicator through health program or policy interventions;
- The health indicator reflected an emerging issue in public health;
- There was evidence that an increase in prevalence or incidence in the indicator is deemed negative to the public’s health;
- The health indicator could be meaningfully measured;
- The health indicator was acceptable as a measure of the underlying characteristic;
- Data to measure the health indicator were available and considered timely.
Sources of Data
Data for each health indicator included in the State of the State’s Health Report were gathered from the best available sources. Mortality data for the demographic variables and county level estimates were acquired from OK2SHARE, the web-based data query system of the Oklahoma State Department of Health. Current demographic data represent deaths for calendar year, while county level data reflect a three-year period. County-level infant mortality rates reflect a five year period. National and state-level mortality data were taken from the Centers for Disease Control and Prevention (CDC) WONDER web-based data query system. Age-adjusted rates using the 2000 US Standard Population were reported (exception: infant mortality).
Prevalence data for diabetes, current asthma, high blood pressure, high cholesterol, and depression were drawn from the Oklahoma Behavioral Risk Factor Surveillance System (BRFSS). It is important to note the BRFSS implemented two methodological changes beginning in data collection year 2011. To adjust to the rapid rise of cellular telephone households and to maintain survey coverage and validity, the BRFSS incorporated cellular telephones to their samples. In addition, the CDC incorporated a new weighting method called “raking” in order to account for declining response rates and differences between the demographic characteristics of respondents and the target population.
County-level data for 2011 forward were estimated using a generalized linear mixed effects regression model with binomial outcome and a logit link function. These models were based on work by Serbotnjak et al., Zhang, X. et al., and Akcin, H. Individual fixed effects included: age group (15 groups), sex (if model was not stratified by sex), race/ethnicity (5 groups), educational attainment, marital status and residence status (own or rent). Random effects included: county of residence and year. Rescaled BRFSS final weights were also included in the models. Modeled county level estimates were adjusted proportionally using state level modeled estimates and direct statewide estimates. Note, the models used in this report are an updated version of the models used in the 2014 State of the State’s Health Report, as a result estimates from this report will not match those that were released in the previous State of the State’s Health Report.
The BRFSS is the source for data documenting behavioral risk patterns. National and state-level data were queried from the CDC BRFSS data system. This includes data for fruit and vegetable consumption; physical activity; current smoking prevalence; obesity; influenza and pneumonia vaccinations among seniors (ages 65 and older); days of limited activity and poor mental and physical health days; self-health rating; dental visitation; alcohol consumption; usual source of care; and lack of health care coverage. Demographic, historic, and county-level data were calculated using SAS.
The Oklahoma Central Cancer Registry provided cancer incidence data (all sites) and site-specific (colon, invasive breast, lung, and prostate) cancer incidence rates. Current demographic data represent a single year and current county-level data reflect three year rates. Data for the United States and the 50 states were acquired through CDC WONDER.
Data for childhood and adolescent immunization rates were obtained from the National Immunization Survey (NIS).
Natality data reported for the demographics and counties were drawn from the Oklahoma birth certificate registry. These data reflect the teenage birth rate for ages 15-17 years, the percentage of births weighing less than 2,500 grams (low birth weight), and the percentage of births occurring to Oklahoma women receiving prenatal care beginning in the first trimester of pregnancy. Demographic and regional data were presented for a single calendar year while county-level data were three year rates. National and state level comparative data were drawn from CDC Wonder. It is important to note that Oklahoma implemented a major revision in how prenatal care data is collected on the Oklahoma Birth Certificate in 2009. Therefore, data in this report cannot be directly compared to earlier years.
Demographic data documenting the percent of people living in poverty reflect data obtained from the American Community Survey (ACS). Region and county-level data reflect data obtained from the Small Area Income and Poverty Estimates Program (SAIPE).
To assign grades to each of the health indicators included in the State of the State’s Health Report, we developed grading scales using the following methods. For each indicator, we examined the U.S. rate and the distribution of rates for the 50 states and the District of Columbia. We calculated the standard deviation for each national rate using the variability of the respective state rates. We assigned cutoff points for each grade level using the standard deviations. Rates ranging between (0.5) standard deviations below the national rate to (0.5) standard deviations above the national rate were assigned the letter grade C (average).
For indicator rates in which higher rates were deemed favorable, rates that were between (0.5) standard deviations and (1.5) standard deviations above the national rate were assigned the letter grade B. Rates that were beyond the (+1.5) standard deviations of the national rate were given the letter grade A. Rates that were (-0.5) and (-1.5) standard deviations below the national rate were given a letter grade of D. A letter grade of F was assigned to grades falling below (-1.5) standard deviations from the national rate. In this situation, the highest (best) rates – those greater than (1.5) standard deviations above the U.S. rate – were assigned As and the lowest (worst) rates – those greater than (1.5) standard deviations below the U.S. rate – were assigned Fs.
For indicator rates in which higher rates were deemed negative, the grading was reversed. That is, rates that were between (0.5) standard deviations and (1.5) standard deviations below the national rate were assigned the letter grade B. Rates that were beyond (-1.5) standard deviations of the national rate were given the letter grade A. Rates above the national rate were given a letter grade of D if the rate was between (+0.5) and (+1.5) standard deviations of the national rate. A letter grade of F was assigned to grades beyond (+1.5) standard deviations of the national rate. Thus, the highest (worst) rates – those greater than (+1.5) standard deviations above the U.S. rate – were assigned Fs and the lowest (best) rates – those greater than 1.5 standard deviations below the U.S. rate – were assigned As.
The grading scheme yields a single distinct scale for each health indicator in the report. Letter grade cutoff points are determined by variability in state level data for each indicator. The grading scales are used to assign grades to select population demographics (e.g., age group, racial/ethnic group, income and education levels), geographic units (e.g., Oklahoma regions and counties, best and worst state rates), and historical trend data.
Limitations of Data
When fewer than 20 events occur in a given county or among a demographic group, the resulting rate is considered unstable or unreliable due to its large relative standard error. This is also the case when making estimates about the population using sample sizes smaller than 50 (as is the case with the BRFSS data). Thus, data for each indicator may not be available for every demographic and county.
Differences in grading occur among groups (i.e., the 18-24 age group may receive a letter grade of A, while the 25-34 age group may receive a letter grade of B on a selected health indicator). This finding does not necessarily indicate a statistically significant difference between the two age groups. No significance testing was done in the completion of this report. Letter grades were assigned, as described above, for the purposes of making relative comparisons for select population subgroups and domains. A difference in assigned letter grade does not denote a significantly worse or better statistical finding, though the finding may suggest a difference of practical importance.
Grades are assigned and comparisons are made among groups using a single distinct grading scale for each indicator. These scales were determined using state-level data and are not specific to a group. For example, the same scale is used to assign a grade for males’ total mortality rate, females’ total mortality rate, Hispanics’ total mortality rate, and the mortality rate among those aged 45-54 years. Males’ total mortality is not being compared to the mortality of males only across the United States, but rather to all mortality in the nation.
The source for a number of health indicators was a surveillance system in which data were collected as part of a survey (e.g., BRFSS). Survey data are subject to sampling error. As a result, responses obtained from the selected sample may differ from the targeted population from which the sample is drawn. It is worthwhile to recognize that a margin of error in sample estimates exists and may impact the distribution of survey responses. This will in turn affect the relative grades of population subgroups. Year-on-year differences may also occur. Rather than representing real changes in the population, yearly fluctuation may indicate sampling error.
Registry data was the source for some health indicators. While these data are not subject to sampling error, health indicator values may fluctuate year-to-year due to small differences in the number of events (i.e., the number of infant deaths per year). This variability may be due to small yearly changes in the number of the underlying events rather than an indication of any meaningful trend.
Mortality-specific Data Concerns
Age. There is a worsening trend related to advancing age given the natural risk of dying as age increases.
Race/Mortality. Race is not self-reported on Death Certificates, and as such is subject to racial misclassification. Oklahoma linkage studies with Indian Health Services indicate one-third of Native American (NA) deaths in Oklahoma are classified as white. Consequently, often NA mortality rates are based on numerators that have been undercounted. Certain Causes of Death that typically are included in NA studies, such as diabetes, tend to have more accurate coding, but will still be under represented.
Hispanics Death Rates. There may be a cultural effect resulting in uncharacteristically low Cause of Death rates. This may be due to the immigrant population returning to their country of birth prior to death. This will underestimate the overall rate of death generally, but particularly among that migrant population group.
Srebotnjak et al.: A novel framework for validating and applying standardized small area measurement strategies, Population Health Metrics 2010 8:26.
Zhang X., Holt J, Lu, H., Multilevel Small Area Estimation for BRFSS Outcomes, EAPO/DEAMPH Analytic Methods Development Team Seminar, Atlanta, GA, April 19, 2012.
Akcin, H., Small Area Estimation (SAE) Project, BRFSS Conference, March 27, 2013.
Martin JA, Hamilton BE, Ventura SJ, Osterman MJK, Matthews, TJ. (2013) Births: Final Data for 2011. National Center for Health Statistics, vol. 62, no. 1.
Centers for Disease Control and Prevention. State-Specific Healthy Life Expectancy at Age 65 Years — United States, 2007–2009. MMWR 2013;62:561-566.