Healthcare Analytics

Research the peer-reviewed literature (attached) to find one application of descriptive analytics in health care (that is: how is descriptive analytics applied in healthcare?). Present this application and explain how it ...

Research the peer-reviewed literature (attached) to find one application of descriptive analytics in health care (that is: how is descriptive analytics applied in healthcare?). Present this application and explain how it can improve healthcare operations and delivery. Please cite your sources in APA Descriptive Analytics Introduction In the first session of the course, analytics was classified as descriptive, predictive, or prescriptive based on the questions asked and the methodologies used to answer those questions. This section will explore descriptive analytics and, more specifically, will review the statistical measures and methods used for summarizing data and examine case studies to provide a high-level understanding of descriptive analytics application. With the SAS activity, students will learn how to code with SAS. ________________________________________ Statistics Most if not all of the methods and techniques used in descriptive analytics applications to analyze data originate from statistics, which can be classified as descriptive or inferential. Descriptive statistics uses formulas and numerical aggregations to summarize a sample of data, describe their essential characteristics, and explain or present the data in a meaningful manner using aggregate numbers, tables, or figures. Note that our analysis is on sample data (data on hand) only. That is, we cannot generalize our results and draw conclusions, inferences, or extrapolations for the entire population. For the data analyst, descriptive statistics serves as a data characterization and validation toolkit that assists in identifying error values, missing values, outliers, distribution anomalies, etc., for the sample data, one variable at a time. The methods used in descriptive statistics are classified as measures of central tendency or as measures of dispersion. Central tendency measures are used to estimate the central positioning of a variable of interest within the data. These measures include the arithmetic mean, median, and mode. See the following table for definitions, as well as Chapter 9 of the textbook. Note that we calculate descriptive statistics for each variable separately. Click here to download a larger view of the table below. Dispersion measures are used to estimate the degree of variation in a given variable of interest. These measures include the range, variance, standard deviation, quartiles, and quartile distance. See the following table for definitions, as well as Chapter 9 of the textbook. Click here to download a larger view of the table below. Descriptive Statistics Graphs Box Plot In addition to the measures described above, there are two graphs that an analyst should review when performing descriptive statistics. First, the box-and-whiskers plot (seen in the following figure), which represents a graphical illustration of several of the statistics described earlier. With a box-and-whiskers plot, one can quickly and visually identify the range, min, max, quartiles, median, and mean values of a variable in a dataset, as well as detect outliers. Click here to download a larger view of the graph below. ________________________________________ Distribution (Histogram) The second plot is the variable frequency distribution (seen in the following figure), which is the frequency of data points counted and plotted over a small number of class labels or numerical ranges. Because many of the statistical methodologies that will be used later in this program adhere to the assumption that variables have normally distributed values – perfectly symmetric on both sides – a histogram can be used to visually inspect whether or not this assumption is valid. Click here to download a larger view of the graph below. When analyzing a variable’s distribution, two interesting measures give us valuable information regarding the shape characteristics of the distribution: skewness and kurtosis. Skewness, a positive or negative value, is used to determine whether the distribution sways to the left or the right, as compared to the perfectly symmetrical normal distribution (see the following figure). Positive skewness implies that the distribution sways to the left (c), while negative skewness implies that the distribution sways to the right (d). If we know the value of skewness for a variable, we can perform appropriate data transformations to correct it or choose appropriate methods to analyze the variable. Click here to download a larger view of the chart below. Kurtosis, a positive or negative value, is used to characterize the peak of the distribution as compared to the peak of a normal distribution. A positive kurtosis indicates a peak taller than the peak of a normal distribution, while a negative kurtosis indicates a shorter peak (see the following figure). As with skewness, knowing the value of kurtosis, we can transform the data to correct it or choose appropriate methods to analyze the variable. Click here to download a larger view of the chart below. All of the measures and graphs described can be computed easily by most statistical software, including SAS. ________________________________________ Inferential Statistics Inferential statistics focuses on drawing inferences or conclusions about the characteristics of a population, not just the sample data. Within inferential statistics, there are techniques for performing hypothesis testing, regression, and correlation. This course will not focus on inferential statistics; however, here is some general information. • Hypothesis testing is used to compare data from different populations (e.g., the average length of stay in hospital A vs. the average length of stay in hospital B) and determine whether the similarities/differences observed are due to chance or due to actual similarities/differences between the data • Regression can be used in descriptive analytics to investigate relationships between different variables (e.g., smoking and cancer), and in predictive analytics to predict events based on one or more explanatory variables (e.g., predicting if one will develop lung cancer if he/she smokes) • Correlation is used to estimate the degree of association between two or more variables ________________________________________ Conclusion This section reviewed some of the statistical measures and plots typically used to describe and summarize data in descriptive analytics applications. While we will revisit most of these in later courses, for this course, students should know what the mean, median, variance, and standard deviation of a variable is, the definitions of kurtosis and skewness, and why we use box plots and histograms. The rest of the section will provide sample cases of descriptive analytics. ________________________________________ Cases Case 1: Underlying Reasons Associated With Hospital Readmission Following Surgery in the United States Paper available here (Links to an external site.)Links to an external site.. The goal of this study was to characterize the reasons for surgical readmissions. Data from 498,875 operations were examined, and researchers found that the unplanned readmission rate was 5.7 percent. The most common reason for readmissions after surgery was surgical site infection. Only a few of the readmissions were related to the chief complaint before surgery. The paper provides additional descriptive statistics for readmission reasons, timing, and factors grouped by the type of surgery. Case 2: Nationwide Analysis of Common Characteristics in OIG Home Health Fraud Cases Report available here (Links to an external site.)Links to an external site.. The report examined data from Medicare claims submitted by home health agencies and physicians to identify fraud. The five criteria used to determine fraud were: 1. High percentage of episodes for which the beneficiary had no recent visits with the supervising physician. 2. High percentage of episodes that were not preceded by a hospital or nursing home stay. 3. High percentage of episodes with a primary diagnosis of diabetes or hypertension. 4. High percentage of beneficiaries with claims from multiple home health agencies. 5. High percentage of beneficiaries with multiple home health readmissions in a short period of time. The study found that 562 home health agencies and 4,502 physicians satisfied at least two of the criteria, and they were flagged for further investigation. Case 3: Automated Detection and Classification of Type 1 vs. Type 2 Diabetes Using Electronic Health Record Data Paper available here (Links to an external site.)Links to an external site.. The study hypothesized that we can use patient data currently residing in electronic health records to detect cases of diabetes and classify its type. This is important because type 2 diabetes can remain undetected in everyday clinical encounters; therefore, there are patients with the diagnosis who may be unaware of it. Researchers used data from approximately 700,000 patients on which they applied the criteria for diabetes detection by the American Diabetes Association to identify patients with diabetes. The criteria used were: 1. Hemoglobin 5 percent 2. Fasting glucose 126mg/dL 3. Prescription for insulin 4. Diabetes diagnosis 5. List of specific medications In total, 43,177 patients exhibited one of the criteria listed above and were flagged as diabetic. Additional rules were implemented to classify between type 1 and type 2 diabetes. Similar rule-based approaches have been used to detect and classify lung and cervical cancer, cardiovascular disease, hypertension, eligibility for clinical trials, bloodstream infections, etc.