7.1 Healthcare Data Types & Scales of Measurement

Healthcare Data Types & Scales of Measurement

In health data analytics, data integrity and analytical validity depend on properly recognizing the fundamental nature of the variables under study. Healthcare systems generate an exceptionally diverse array of data—ranging from qualitative diagnostic descriptions and categorical clinical classifications to discrete physiological event counts and high-precision continuous laboratory measurements. For a Certified Health Data Analyst (CHDA), understanding the typology of data and their respective scales of measurement is not merely a theoretical exercise; it dictates which mathematical operations are valid, which descriptive summary statistics must be reported, and which inferential statistical methodologies can be legitimately applied.

Applying an inappropriate mathematical operation to a variable—such as calculating the arithmetic mean of an ordinal cancer stage or averaging patient satisfaction categories—introduces severe methodological distortion and can lead to erroneous clinical, financial, or operational conclusions.

---

1. Stevens' Typology: The Four Scales of Measurement

In 1946, psychologist and psychophysicist Stanley Smith Stevens introduced a foundational classification system that categorizes all data into four hierarchical levels of measurement: Nominal, Ordinal, Interval, and Ratio. Each level builds upon the properties of the preceding scale, adding greater mathematical structure and permitting a broader suite of analytical operations.

+---------------------------------------------------------------------------------------------------+
|                         STEVENS' HIERARCHY OF MEASUREMENT SCALES                                  |
+---------------------------------------------------------------------------------------------------+
                                                  |
  [1. NOMINAL]  --> Identity / Classification Only (No order, no distance, no true zero)
                    Examples: Blood Type (A, B, AB, O), ICD-10-CM Codes, Patient Biological Sex
                                                  |
  [2. ORDINAL]  --> Identity + Meaningful Rank Order (Unequal or unknown intervals, no true zero)
                    Examples: Cancer Stage (0-IV), ASA Score (I-VI), CAHPS Survey Likert Scales
                                                  |
  [3. INTERVAL] --> Identity + Rank Order + Equal Measured Intervals (Arbitrary zero point)
                    Examples: Body Temperature (°F / °C), Calendar Dates, Standardized Test Scores
                                                  |
  [4. RATIO]    --> Identity + Rank Order + Equal Intervals + Absolute True Zero Point
                    Examples: Length of Stay (days), Total Hospital Costs ($), Blood Pressure (mmHg)
+---------------------------------------------------------------------------------------------------+

1. Nominal Scale (Categorical / Classification Scale)

The nominal scale represents the lowest level of measurement. Data at this level consist of mutually exclusive and collectively exhaustive categories. Numbers may be assigned to categories purely as arbitrary labels or identifiers, but they carry no inherent quantitative value, rank, or mathematical order.

• Defining Characteristics: Qualitative, unordered, identity-only relationship ($A = B$ or $A \neq B$).
• Healthcare Examples:
  • Patient Demographics: Biological sex (1 = Male, 2 = Female), Race/Ethnicity categories, Marital status.
  • Clinical Classifications: ABO Blood Type (A, B, AB, O), Rh Factor (Positive, Negative), ICD-10-CM diagnosis codes (e.g., E11.9 for Type 2 diabetes mellitus without complications vs. I10 for Essential hypertension).
  • Operational / Administrative Codes: Admission Source (1 = Emergency Room, 2 = Physician Referral, 3 = Transfer from another hospital), Discharge Disposition (01 = Discharged to home, 02 = Discharged/transferred to short-term general hospital, 20 = Expired), Attending Provider ID, Health Insurance Payer (Medicare, Medicaid, Commercial, Self-Pay).
• Permissible Mathematical & Statistical Operations:
  • Counting frequencies (absolute counts $n$ and relative percentages %).
  • Mode (the most frequently occurring category) is the only valid measure of central tendency.
  • Non-parametric contingency analysis: Cross-tabulations, Chi-Square ($\chi^2$) Goodness-of-Fit and Test of Independence, Fisher's Exact Test, Cramér's $V$.
  • Prohibited Operations: Mean, median, addition, subtraction, multiplication, division, standard deviation, and ranking.

2. Ordinal Scale (Rank-Ordered Scale)

The ordinal scale captures data where categories exhibit a clear, meaningful natural order or hierarchy. However, the mathematical distance (interval) between successive categories is neither constant nor quantifiable. While we know that Rank 3 is greater than Rank 2, we cannot assume that the difference between Rank 3 and Rank 2 is identical to the difference between Rank 2 and Rank 1.

• Defining Characteristics: Identity + directional rank ordering ($A > B$, $A = B$, or $A < B$), with unequal or unknown intervals and no true zero.
• Healthcare Examples:
  • Clinical Staging & Acuity: Cancer Staging (Stage 0, Stage I, Stage II, Stage III, Stage IV), New York Heart Association (NYHA) Functional Classification for heart failure (Class I, II, III, IV), American Society of Anesthesiologists (ASA) Physical Status Classification (ASA I through ASA VI), Emergency Severity Index (ESI Levels 1–5).
  • Functional Assessment Scales: Glasgow Coma Scale (GCS, 3–15), Braden Scale for predicting pressure ulcer risk (6–23), Rankin Scale for stroke disability (0–6).
  • Patient Experience & Pain Ratings: HCAHPS Likert survey responses (1 = Never, 2 = Sometimes, 3 = Usually, 4 = Always), Visual Analog Pain Scale ratings (0 = No pain to 10 = Worst imaginable pain).
• Permissible Mathematical & Statistical Operations:
  • Frequency counts and cumulative percentage distributions.
  • Median (50th percentile) is the primary permissible measure of central tendency; Interquartile Range (IQR) and percentiles measure dispersion.
  • Rank-order correlation: Spearman's rank correlation coefficient ($\rho$), Kendall's tau ($\tau$).
  • Non-parametric hypothesis testing: Mann-Whitney $U$ test, Wilcoxon signed-rank test, Kruskal-Wallis $H$ test.
  • Prohibited Operations: Arithmetic mean, standard deviation, addition, subtraction, ratios. (Note: While calculating the "mean" of a 5-point Likert scale or pain score is frequently observed in informal healthcare reporting, it violates strict mathematical typology because the psychological distance between survey increments is not equal).

3. Interval Scale (Equal Interval Scale)

The interval scale possesses ordered categories with equal, standardized units of measurement throughout the entire continuum. The difference between 10 and 20 units is mathematically identical to the difference between 30 and 40 units. However, interval scales lack an absolute, non-arbitrary true zero point. Zero on an interval scale is simply an arbitrary reference point and does not represent the complete absence of the attribute being measured.

• Defining Characteristics: Identity + Rank Order + Equal Intervals ($A - B = C - D$), but arbitrary zero point ($A / B$ is meaningless).
• Healthcare Examples:
  • Temperature: Body temperature measured in Fahrenheit (°F) or Celsius (°C). Zero degrees (0°F or 0°C) does not represent the total absence of thermal energy. Consequently, an analyst cannot state that a patient with a fever of 104°F is "twice as hot" as a patient at 52°F.
  • Chronological Timestamps / Calendar Dates: Admission date (e.g., August 23, 2026), Julian day of the year. The interval between August 1 and August 5 is exactly 4 days, equal to the interval between August 10 and August 14, but "Year 0" is an arbitrary chronological convention.
  • Standardized Psychometric & Intelligence Scores: IQ scores, standardized cognitive assessment test batteries with standardized normalized score distributions.
• Permissible Mathematical & Statistical Operations:
  • Addition and subtraction (calculating differences, elapsed days, change scores $\Delta = x_2 - x_1$).
  • Arithmetic Mean, Standard Deviation, Variance, and Range.
  • Parametric correlation and regression: Pearson correlation coefficient ($r$), linear regression, Student's $t$-test, Analysis of Variance (ANOVA).
  • Prohibited Operations: Multiplication or division of raw scale values, calculating ratios (e.g., 100°C is not $2\times$ 50°C), Geometric Mean, and Coefficient of Variation.

4. Ratio Scale (Absolute True Zero Scale)

The ratio scale represents the highest level of measurement. It encompasses all the properties of nominal, ordinal, and interval scales, with the addition of an absolute, non-arbitrary true zero point that signifies the complete absence of the physical quantity or attribute being measured.

• Defining Characteristics: Identity + Rank Order + Equal Intervals + Absolute True Zero ($A / B$ is mathematically valid).
• Healthcare Examples:
  • Temporal Measures: Inpatient Length of Stay (LOS in days or hours; 0 days = same-day outpatient discharge; an 8-day stay is exactly twice as long as a 4-day stay), ED wait time in minutes, operating room turnaround time.
  • Physical & Physiological Measures: Patient age in years/months, body weight in kilograms, height in centimeters, systolic and diastolic blood pressure in mmHg (0 mmHg represents complete absence of hydrostatic pressure), respiratory rate (breaths/min), heart rate (beats/min).
  • Clinical Laboratory Concentrations: Serum creatinine (mg/dL), Blood Urea Nitrogen (BUN, mg/dL), Fasting blood glucose (mg/dL), Hemoglobin A1c (%), White blood cell count ($10^3/\mu\text{L}$), Plasma viral load (copies/mL).
  • Financial & Operational Measures: Total hospital billed charges ($), direct variable cost of surgery ($), reimbursement payment amount ($), nurse staffing hours per patient day (HPPD), units of packed red blood cells transfused.
• Permissible Mathematical & Statistical Operations:
  • All mathematical operations: addition, subtraction, multiplication, division, exponentiation, logarithms.
  • All central tendency and dispersion metrics: Arithmetic Mean, Median, Mode, Geometric Mean (vital for log-normal viral titers), Harmonic Mean, Standard Deviation, Variance, IQR, and Coefficient of Variation (CV).
  • All advanced parametric modeling: Pearson correlation, generalized linear models, multiple regression, non-linear survival modeling (Cox proportional hazards), time-series forecasting.

---

2. Measurement Scales Comparison Matrix

The following matrix synthesizes the structural properties, valid mathematical operations, descriptive metrics, and healthcare examples across all four measurement scales:

---

3. Qualitative vs. Quantitative and Discrete vs. Continuous Variables

Beyond Stevens' scales, health data analysts classify variables along structural and numerical dimensions:

+---------------------------------------------------------------------------------------------------+
|                             HEALTHCARE VARIABLE CLASSIFICATION TAXONOMY                           |
+---------------------------------------------------------------------------------------------------+
                                                  |
                 +--------------------------------+--------------------------------+
                 |                                                                 |
       [QUALITATIVE (CATEGORICAL)]                                       [QUANTITATIVE (NUMERICAL)]
                 |                                                                 |
      +----------+----------+                                           +----------+----------+
      |                     |                                           |                     |
 [NOMINAL]              [ORDINAL]                                  [DISCRETE]            [CONTINUOUS]
 - Unordered labels     - Ranked order                             - Count data (integers)- Infinite continuum
 - ICD-10 codes         - Cancer staging (I-IV)                    - Readmissions count   - Hemoglobin A1c (%)
 - Biological sex       - Pain score (0-10)                        - Surgical unit beds   - Serum creatinine
 - Blood type           - CAHPS survey scale                       - Number of ED visits  - Blood pressure

Qualitative (Categorical) Variables

Qualitative variables describe attributes, qualities, or characteristics that fall into distinct groups or classes. They cannot be measured on a continuum.

1. Binary / Dichotomous Variables: Categorical variables with exactly two mutually exclusive states.
   • Examples: In-hospital Mortality (0 = Alive, 1 = Deceased), 30-Day Readmission (0 = No, 1 = Yes), Surgical Site Infection (0 = Absent, 1 = Present), Present on Admission flag (Y = Yes, N = No).
2. Polytomous / Multiclass Variables: Categorical variables with three or more distinct classes.
   • Examples: Marital Status (Single, Married, Divorced, Widowed), Primary Health Insurance Carrier (Medicare Fee-for-Service, Medicare Advantage, Medicaid, Commercial PPO, Self-Pay).

Quantitative (Numerical) Variables

Quantitative variables represent measurable numerical quantities where numbers reflect actual amounts, counts, or magnitudes.

1. Discrete Quantitative Variables: Variables resulting from a counting process that can take on only distinct, separate, non-negative integer values. There are no intermediate values between consecutive integers.
   • Mathematical Property: Countable set of values ${0, 1, 2, 3, \dots}$.
   • Healthcare Examples: Number of acute emergency department visits in the past 12 months, count of active prescription medications on the medication reconciliation flowsheet, number of staffed ICU beds, count of secondary chronic comorbidities, total number of prior caesarean sections.
   • Analytical Handling: Modeled using Poisson distribution or Negative Binomial regression when dealing with overdispersed event counts.
2. Continuous Quantitative Variables: Variables resulting from a physical or physiological measurement process that can assume an infinite number of possible real-number values along a specified continuous interval. The precision of the measurement is limited only by the sensitivity of the measuring device or laboratory analyzer.
   • Mathematical Property: Real numbers $\mathbb{R}$ within a given range $[a, b]$.
   • Healthcare Examples: Serum Potassium concentration ($4.18\text{ mmol/L}$), Fasting Blood Glucose ($108.4\text{ mg/dL}$), Hemoglobin A1c ($7.2%$), Patient weight ($78.45\text{ kg}$), Elapsed surgical duration ($142.6\text{ minutes}$).
   • Analytical Handling: Evaluated using parametric descriptive statistics (mean, variance) or non-parametric alternatives when distributions deviate from normality.

---

4. Healthcare Data Distributions & Shapes

Healthcare phenomena rarely follow idealized statistical distributions. A health data analyst must examine the empirical shape of a variable's distribution before selecting descriptive summaries or predictive models.

+---------------------------------------------------------------------------------------------------+
|                             COMMON HEALTHCARE DISTRIBUTION SHAPES                                 |
+-----------------------------------+-----------------------------------+---------------------------+
| NORMAL (GAUSSIAN) DISTRIBUTION    | POSITIVE (RIGHT) SKEWED           | NEGATIVE (LEFT) SKEWED    |
| - Symmetric bell curve            | - Long right tail (high outliers) | - Long left tail          |
| - Mean = Median = Mode            | - Mode < Median < Mean            | - Mean < Median < Mode    |
| - Physiological reference vitals  | - Inpatient LOS, Total Charges    | - Gestational Age at birth|
+-----------------------------------+-----------------------------------+---------------------------+

1. Normal (Gaussian) Distribution

• Shape: Perfectly symmetrical, bell-shaped distribution where values taper off equally on both sides of the center. The asymptotic tails extend toward infinity but never touch the horizontal axis.
• Relationship of Averages: $\text{Mean} = \text{Median} = \text{Mode}$.
• Healthcare Examples: Physiological biomarkers in healthy, non-diseased baseline populations (e.g., adult height, serum sodium in ambulatory outpatients, birth weight among healthy full-term singleton infants).
• Analyst Rule: Standard parametric statistics (Mean, Standard Deviation, $t$-tests, ANOVA) are fully valid.

2. Positive (Right) Skewed Distribution

• Shape: The mass of the distribution is concentrated at lower values on the left, while a long, extended tail stretches out toward extreme high values on the right.
• Relationship of Averages: $\text{Mode} < \text{Median} < \text{Mean}$. The arithmetic mean is heavily pulled upward into the right tail by extreme high outliers, making it an unrepresentative measure of central tendency for typical patients.
• Healthcare Examples: Ubiquitous across healthcare operations and finance:
  • Inpatient Length of Stay (LOS): Most acute care patients are discharged within 2 to 5 days, but a small subset of catastrophic ICU or burn patients remain hospitalized for 30 to 120 days, pulling the mean LOS far above the median.
  • Total Hospital Charges & Direct Costs: Most encounters generate modest costs ($3,000–$12,000), but multi-organ failure or ECMO cases generate charges exceeding $500,000.
  • Emergency Department Wait Times & Diagnostic Turnaround Times.
• Analyst Rule: Never report the arithmetic mean alone for right-skewed clinical or cost data. Always report the Median and Interquartile Range (IQR), or perform logarithmic transformations ($y = \ln(x)$) before parametric modeling.

3. Negative (Left) Skewed Distribution

• Shape: The mass of the distribution is concentrated at higher values on the right, while an elongated tail stretches out toward extreme low values on the left.
• Relationship of Averages: $\text{Mean} < \text{Median} < \text{Mode}$. The arithmetic mean is pulled downward by extreme low-value outliers.
• Healthcare Examples:
  • Gestational Age at Delivery: The vast majority of births occur between 37 and 41 completed weeks of gestation (the peak/mode), while a long tail of premature births extends downward to 23–36 weeks.
  • Age at Diagnosis for Adult Chronic Conditions: Diagnoses of chronic diseases like Alzheimer's disease or osteoarthritis cluster heavily among elderly cohorts (ages 75–85), with a long tail extending downward to early-onset cases in younger adults.
  • Patient Experience / Satisfaction Scores: Standard CAHPS survey questions frequently exhibit extreme negative skew (ceiling effect), where 85–90% of respondents select top-box scores (Always or 10/10), with a sparse tail of dissatisfied ratings.

4. Bimodal & Multimodal Distributions

• Shape: Distributions exhibiting two (bimodal) or more (multimodal) distinct local peaks (modes), indicating the presence of distinct underlying sub-populations combined within a single dataset.
• Healthcare Examples:
  • Hospital Emergency Department Arrivals by Hour of Day: Peaks typically occur during the late morning surge (10:00 AM – 12:00 PM) and the evening post-work surge (6:00 PM – 8:00 PM), with deep troughs at 4:00 AM.
  • Age Distribution in a General Community Hospital ED: Bimodal peaks representing pediatric patients (infants and young children with viral febrile illnesses) and geriatric patients (ages 65+ with multi-morbid decompensation), with lower utilization among young adults (ages 18–30).
  • Systolic Blood Pressure in Mixed Populations: Distinct modes reflecting normotensive individuals and untreated hypertensive patients.
• Analyst Rule: Bimodal distributions should never be summarized with a single measure of central tendency. The analyst must stratify the dataset and report separate descriptive statistics for each distinct sub-cohort.