Certified Health Data Analyst (CHDA) Practice Review
Healthcare Analytics Taxonomy and Applications
Descriptive Analytics:
Definition: Descriptive analytics summarizes, aggregates, and describes historical data to answer the core operational question: What happened?
Reporting of Historical Performance: A monthly management report summarizing coder productivity, overall denial rate, and Case Mix Index () for the prior month is an example of descriptive analytics because it consolidates completed operational events.
Specialty Dashboards: Displaying the previous quarter's claim denial rates stratified by clinical specialty constitutes descriptive reporting by visualizing past revenue cycle outcomes.
Inpatient Metrics: Calculating and publishing an average length of stay () of for the preceding month summarizes past inpatient utilization.
Diagnostic Analytics:
Definition: Diagnostic analytics evaluates historical data patterns, relationships, and anomalies to identify root causes and address the question: Why did it happen?
Workflow Change Investigation: Investigating root causes when claim denial rates spike following an electronic health record () workflow update represents diagnostic analysis.
Length of Stay Anomaly Isolation: Analyzing inpatient data to determine that an observed increase in average was concentrated specifically among patients awaiting placement in skilled nursing facilities () exemplifies diagnostic analysis by isolating the driver of the variance.
Predictive Analytics:
Definition: Predictive analytics utilizes statistical modeling, machine learning algorithms, and historical data patterns to estimate future outcomes, answering: What is likely to happen?
Readmission Modeling: A statistical model that assesses clinical variables to estimate which inpatients are likely to experience an unplanned hospital readmission within of discharge.
Emergency Department Demand Forecasting: Building a time-series or regression model to forecast the total patient volume arriving at an emergency department () during the subsequent month.
Denial Propensity Scoring: Identifying clinical, demographic, and billing characteristics strongly associated with past payer rejections to predict which newly submitted claims will encounter denials.
Separation of Extraction and Modeling: Extracting candidate predictor fields (such as patient age, principal diagnosis, count of prior hospital admissions, and historical length of stay) from an underlying relational database via Structured Query Language () represents the preliminary data preparation step; generating risk scores from these features requires a distinct analytic modeling step.
Prescriptive Analytics:
Definition: Prescriptive analytics applies optimization formulas, simulation methods, and decision rules to determine the optimal course of action, answering: What action should be taken?
Clinical Intervention Assignment: An automated clinical decision support system that evaluates a high-risk patient's profile and prescribes the specific post-discharge care management protocol to be initiated.
Staffing Optimization: An automated scheduling system that translates predicted emergency department patient arrival volumes into the optimal number of nurses to schedule across specific operational shifts.
Constrained Risk Mitigation: Coupling a predictive readmission model with a mathematical optimization algorithm to select the specific intervention bundle that minimizes readmission probability at the lowest total implementation cost.
SQL Syntax, Relational Algebra, and Query Logic
Fundamental Clauses and Execution Roles:
SELECT: Identifies the specific data attributes, mathematical calculations, aggregate expressions, or database columns to return in the query output.FROM: Specifies the source table, view, or joined entities from which records are retrieved.WHERE: Defines the row-level filtering criteria that each record must evaluate to true in order to be included in the result set.*(Wildcard): Instructs the database engine to return every column present in the specified table schema (e.g.,SELECT * FROM Patients;).
Comparison and Logical Operators:
Logical
AND: Requires every connected logical condition to evaluate to true for a row to be returned. If an analyst queriesWHERE systolic_bp > 194 AND weight > 94, a patient exhibiting a systolic blood pressure of and a weight of is excluded because the second criterion () evaluates to false.Logical
OR: Returns a row if at least one of the connected conditions evaluates to true. For the filterWHERE age > 80 OR systolic_bp > 180, a patient aged with a systolic blood pressure of is included because the blood pressure condition is satisfied ().Strict Inequalities vs. Inclusive Bounds:
The operator
>defines a strict upper boundary that excludes the threshold value itself (e.g.,WHERE los > 5selects stays of or more, omitting stays of exactly ).The operator
>=establishes an inclusive lower boundary that retains the exact threshold value (e.g.,WHERE los >= 5includes stays equal to alongside higher values).Explicit inclusion intervals require two bounded operators joined by logical
AND(e.g., selecting patients from age through inclusive is expressed asWHERE age >= 18 AND age <= 64).Multiple simultaneous criteria requiring threshold adherence must combine explicit conditions (e.g., isolating poorly controlled, obese diabetic patients requires
WHERE a1c > 9.0 AND bmi >= 30).
Aggregate Functions:
AVG(): Computes the arithmetic mean of an assigned numeric column across all qualifying rows (e.g.,SELECT AVG(los) FROM Encounters;).COUNT(*): Calculates the total integer count of rows returned by the query, irrespective of null values in individual columns (e.g.,SELECT COUNT(*) FROM Encounters;).
Relational Joins:
INNER JOIN: Evaluates two related tables and returns only records that have matching keys in both entities. Unmatched records in either table are discarded.LEFT JOIN(Left Outer Join): Preserves every single record from the left table regardless of whether a matching record exists in the right table. For non-matching left records, right-table attributes are populated withNULL.RIGHT JOIN(Right Outer Join): Preserves every record from the right table, retaining records even when no corresponding record exists in the left table, with left-table attributes populated withNULL.FULL OUTER JOIN: Returns all records from both tables, combining matched rows and preserving left-only and right-only records withNULLfill for non-matching sides.
Unmatched Record Detection Pattern:
To identify parent records that lack corresponding child records (such as identifying registered patients who have never completed a clinical encounter), implement a
LEFT JOINfrom the parent to child table, followed by a filter verifying that the primary key of the child table evaluates toNULL:SELECT p.patient_id FROM Patients p LEFT JOIN Encounters e ON p.patient_id = e.patient_id WHERE e.patient_id IS NULL;
Healthcare Query Implementations:
Filtering Categorical Billing Status: Extracting denied claims requires filtering the categorical status string:
SELECT * FROM Claims WHERE claim_status = 'Denied';.Targeted Column Extraction: Retrieving specific denial metadata:
SELECT claim_id, denial_reason FROM Claims WHERE claim_status = 'Denied';.Multi-Attribute Threshold Filtering: Isolating Medicare encounters exceeding an acute length of stay:
SELECT patient_id FROM Encounters WHERE payer = 'Medicare' AND los > 5;.Downstream Risk Score Consumption: Selecting high-risk patients identified by a predictive algorithm where the threshold is or above:
SELECT patient_id FROM RiskScores WHERE risk_score >= 0.80;.
Descriptive Statistics, Inferential Methods, and Data Visualization
Measures of Central Tendency and Dispersion:
Range: The arithmetic difference between the maximum and minimum observed values in a dataset:
For the values :
Mean: The arithmetic average of a dataset containing observations:
For the values :
Median: The middle value in an ordered dataset. For an even number of observations ( is even), calculate the arithmetic mean of the two central numbers:
For the sorted set (), average the 3rd and 4th observations:
Outlier Resistance:
The median is resistant to extreme outliers, whereas the mean, range, and standard deviation are sensitive to extreme skewness.
In a sample of inpatient stays displaying lengths of stay of , the single extreme value of dramatically inflates the mean and range, making the median the preferred, least affected measure of central tendency.
Data Visualization Selection Principles:
Line Graph: The optimal visualization for displaying trends, rates, and patterns across sequentially ordered, continuous time intervals (e.g., tracking monthly coding denial rates from January through December).
Bar Chart: The standard chart type used to compare discrete numeric values or summary statistics across distinct categories, cohorts, or facilities (e.g., comparing average length of stay across four separate hospital facilities).
Histogram: Designed to display the underlying frequency distribution of a single continuous numeric variable by segmenting data into adjacent, non-overlapping intervals or bins (e.g., plotting the frequency distribution of patient ages grouped into age bands).
Scatterplot: The primary visualization used to explore the bivariate relationship, distribution, and potential correlation between two paired continuous numeric variables (e.g., evaluating whether patient length of stay is associated with total hospital charges).
Correlation vs. Causation:
A scatterplot whose data points trend upward from left to right as the horizontal variable increases demonstrates a positive correlation (e.g., older age is associated with an increased length of stay).
Observational statistical correlation indicates co-movement between variables but does not prove a causal relationship.
Statistical Process Control ():
Control charts track process performance over time relative to three lines: a center line (representing the process mean) bounded by an Upper Control Limit () and a Lower Control Limit ().
Inherent random variation within the control limits reflects common-cause variation.
Any data point plotted outside the established control boundaries (such as a monthly value exceeding the ) serves as an operational signal of special-cause variation that warrants diagnostic investigation.
Inferential Statistical Hypothesis Testing:
Two-Sample Independent -Test: Parametric test appropriate for comparing the difference between the arithmetic means of two independent groups (e.g., assessing whether mean differs significantly between two distinct hospitals).
One-Way Analysis of Variance (): Statistical technique used to determine whether statistically significant differences exist among the means of three or more independent groups (e.g., evaluating mean across four separate hospital facilities).
Chi-Square () Test of Independence: Non-parametric test utilized to determine whether a significant association exists between two categorical, nominal, or ordinal variables (e.g., evaluating whether primary payer category is associated with discharge disposition).
Work Relative Value Units (wRVU) and Physician Productivity Analysis
Productivity Frameworks: Volume vs. Clinical Intensity:
Encounter Volume (Visit Count): A raw quantitative count measuring clinical access, patient throughput, and total capacity ().
Work Relative Value Units (): A standardized, objective measure quantifying the relative time, technical skill, physical effort, and mental stress expended by a physician to perform a given service or procedure.
Case Mix Index () and Discharged Not Final Billed (): reflects the relative severity and resource intensity of an inpatient population, while tracks revenue cycle billing lag. Neither metric directly measures individual physician work productivity.
Work Output Primacy: When assessing true clinical productivity defined as total physician work delivered, total represent a more comprehensive metric than gross encounter volume because adjust for service complexity.
Mathematical Calculation of Total wRVUs:
Total physician work generated across diverse Current Procedural Terminology () codes is calculated by taking the sum of the product of visit volume and the assigned weight for each specific code:
Sample Procedure-Level Calculation:
A clinician records the following procedural volume:
at code ( each)
at code ( each)
at code ( each)
at code ( each)
Calculate total work output by summing the individual code contributions:
Comparative Physician Productivity Analysis:
Established Procedural Weights:
CPT :
CPT :
CPT :
CPT :
CPT :
CPT :
CPT :
CPT :
Physician Encounter Distributions:
Dr A: Visits across the eight codes:
Dr B: Visits across the eight codes:
Dr C: Visits across the eight codes:
Total Encounter Volume Comparison:
Dr A:
Dr B:
Dr C:
Result: Dr C achieved the highest encounter volume ().
Total wRVU Output Comparison:
Dr A:
Dr B:
Dr C:
Result: Dr B demonstrated the highest overall productivity when measured by total work ().
Average Work Intensity per Visit:
Dr A:
Dr B:
Dr C:
Result: Dr B produced the highest average work intensity per clinical encounter ().
Evaluation and Management () Distribution Dynamics:
A high proportion of Level-5 visits indicates greater case mix intensity within encounters, but total productivity requires multiplying intensity by total volume.
A practitioner managing lower overall patient volume can outperform higher-volume peers in total work output if their clinical service mix involves higher code intensities (e.g., Dr Y completing and producing generated more physician work than Dr X completing and generating ).