I built the Tableau portion of the project to make the combined CMS dataset exploratory rather than presenting only a fixed set of charts.
The story allows the viewer to move from broad hospital characteristics into more specific questions:
How do ratings differ by hospital ownership?
Do patients who report better communication also tend to rate the hospital more highly?
How closely do patient-experience measures align with CMS Overall Star Ratings?
Are any patient-experience measures meaningfully associated with 30-day readmission rates?
Do those patterns change depending on the condition being examined?
I created four Tableau parameters to allow the same visualizations to answer multiple questions.
Select X-Axis Touchpoint Metric
Controls the patient-experience measure displayed on the X-axis.
Available metrics include:
Nurse Communication (% Always)
Doctor Communication (% Always)
Medication Communication (% Always)
Cleanliness Star Rating
Quietness Star Rating
Discharge Info Star Rating
Select Y-Axis Outcome
Allows the user to compare the selected touchpoint with:
Patient Advocacy of Hospital (% Definitely Recommend)
HCAHPS Summary Star Rating
HCAHPS Overall Star Rating
CMS Overall Star Rating
Select Y-Axis Readmission Rate
Allows the user to select:
Heart Failure Readmission Rate
COPD Readmission Rate
Pneumonia Readmission Rate
AMI Readmission Rate
CABG Readmission Rate
Elective Hip/Knee Surgery Readmission Rate
Y-Axis Selector
Used in the communication-comparison view to switch between Doctor Communication and Medication Communication while keeping Nurse Communication on the X-axis.
Dynamic Readmission Rate
CASE [Select Y-Axis Readmission Rate]
WHEN 'Heart Failure Readmission Rate' THEN [Readm Rate Heart Failure]
WHEN 'COPD Readmission Rate' THEN [Readm Rate Copd]
WHEN 'Pneumonia Readmission Rate' THEN [Readm Rate Pneumonia]
WHEN 'AMI (Acute Heart Attack) Readmission Rate' THEN [Readm Rate Ami]
WHEN 'CABG (Bypass Surgery) Readmission Rate' THEN [Readm Rate Cabg]
WHEN 'Elective Hip/Knee Surgery Readmission Rate' THEN [Readm Rate Hip Knee]
END
This field lets one readmission scatterplot update without creating a separate worksheet for every condition.
Dynamic X-Axis
CASE [Select X-Axis Touchpoint Metric]
WHEN 'Nurse Communication (% Always)' THEN [Nurse Comm Always Pct]
WHEN 'Doctor Communication (% Always)' THEN [Doctor Comm Always Pct]
WHEN 'Medication Communication (% Always)' THEN [Med Comm Always Pct]
WHEN 'Cleanliness Star Rating' THEN [Hcahps Cleanliness Star Rating]
WHEN 'Quietness Star Rating' THEN [Hcahps Quietness Star Rating]
WHEN 'Discharge Info Star Rating' THEN [Hcahps Discharge Info Star Rating]
END
This provides a common X-axis for all six patient-experience measures.
Dynamic Y-Axis Outcome
CASE [Select Y-Axis Outcome]
WHEN 'Patient Advocacy of Hospital (% Definitely Recommend)' THEN [Recommend Definitely Pct]
WHEN 'HCAHPS Summary Star Rating' THEN [Hcahps Summary Star Rating]
WHEN 'HCAHPS Overall Star Rating' THEN [Hcahps Overall Star Rating]
WHEN 'CMS Overall Star Rating' THEN [Cms Overall Star Rating]
END
This lets the viewer compare the same patient touchpoint with different definitions of patient and hospital performance.
Selected Y-Axis Communication %
CASE [Y-Axis Selector]
WHEN 'Doctor Communication (% Always)' THEN [Doctor Comm Always Pct]
WHEN 'Medication Communication (% Always)' THEN [Med Comm Always Pct]
END
This field was used to directly compare communication domains against Nurse Communication.
One visualization compares Nurse Communication with Doctor or Medication Communication.
The Doctor Communication model was:
Doctor Communication % = 0.7119 × Nurse Communication % + 22.31
R² = 0.644
p < 0.0001
The Medication Communication model was:
Medication Communication % = 0.8711 × Nurse Communication % – 7.81
R² = 0.591
p < 0.0001
Both relationships are strongly positive.
The practical takeaway is that hospitals with higher patient ratings for nursing communication also generally receive higher ratings for physician and medication communication. This suggests that communication performance tends to cluster within hospitals rather than behaving as completely independent provider-level measures.
The story also compares CMS and HCAHPS ratings across hospital ownership categories.
Government – Hospital District or Authority facilities had some of the highest average patient-experience ratings in the ownership analysis, while Physician-Owned hospitals were among the lowest on both patient and CMS ratings.
The dashboard also highlights that patient ratings and CMS Overall Star Ratings do not always move together. Some ownership categories show a noticeable difference between how patients rate their experience and the hospital's average CMS Overall Star Rating.
These ownership-level averages should be interpreted with the number of hospitals and missing values in mind. They describe the hospitals available in this dataset rather than proving that ownership structure itself causes performance differences.
Cook County contains the largest number of hospitals in the Illinois dataset.
Within the dashboard analysis, Cook County's average available CMS Overall Star Rating was approximately 2.83 out of 5.
I included this view to provide geographic context rather than treating statewide hospital averages as evenly distributed across Illinois.
Across the statewide exploratory analysis, Heart Failure had the highest average 30-day readmission rate at 19.92%, followed by COPD at 18.39%.
Hip/Knee Replacement had the lowest average readmission rate at 5.06%.
The ownership view showed the same broad pattern: heart failure generally remained the highest-readmission condition, while elective hip/knee procedures were substantially lower.
The parameterized patient-outcome scatterplot produced the strongest relationships in the project.
Nurse Communication had the strongest relationship with HCAHPS Summary Star Rating:
R² = 0.739
It also showed strong relationships with:
HCAHPS Overall Star Rating: R² = 0.652
Definitely Recommend: R² = 0.638
Discharge Information was another strong patient-experience measure:
Discharge Information vs. HCAHPS Summary Star Rating: R² = 0.679
Discharge Information vs. Definitely Recommend: R² = 0.497
Cleanliness, quietness, physician communication, and medication communication also showed moderate-to-strong relationships with the HCAHPS patient-rating outcomes.
The consistent pattern is that individual patient-experience touchpoints are closely related to patients' broader evaluation of their hospital stay.
The same relationships became substantially weaker when CMS Overall Star Rating was selected as the outcome.
For the three provider communication percentages:
Nurse Communication: R² = 0.106
Medication Communication: R² = 0.078
Doctor Communication: R² = 0.046
Discharge Information had the highest R² against CMS Overall Star Rating among the six selected touchpoints:
R² = 0.239
The dashboard therefore demonstrates a useful distinction between measures.
A hospital can receive strong evaluations from patients for communication and other aspects of their experience without those measures explaining most of the variation in its CMS Overall Star Rating.
The final parameterized scatterplot compares all six patient-experience measures against all six readmission outcomes.
This created 36 possible combinations.
The overall result was consistent:
No model had an R² greater than 0.083.
Only four of the 36 models met the p ≤ 0.05 significance threshold.
Doctor Communication vs. Heart Failure Readmission
R² = 0.042
p = 0.018
The regression slope was negative, indicating that hospitals with higher Doctor Communication percentages tended to have slightly lower heart-failure readmission rates.
Doctor Communication vs. Pneumonia Readmission
R² = 0.040
p = 0.015
Again, the relationship was negative but small.
Medication Communication vs. Pneumonia Readmission
R² = 0.040
p = 0.015
Higher Medication Communication percentages were associated with slightly lower pneumonia readmission rates.
Discharge Information vs. COPD Readmission
R² = 0.044
p = 0.028
Hospitals with higher Discharge Information ratings tended to have slightly lower COPD readmission rates.
These findings are statistically significant, but the R² values show that each relationship explains only about 4% of the observed variation.
I therefore do not interpret these models as evidence that improving one communication score by itself would meaningfully reduce readmissions.
I included both p-values and R² values because they answer different questions.
The p-value helps determine whether the observed linear relationship is unlikely to be explained by random variation alone under the model assumptions.
R² describes how much of the variation in the outcome is explained by the fitted linear relationship.
This distinction is especially important in the readmission portion of the project. Several relationships reach statistical significance, but their R² values remain very low. The association may be detectable without being strong enough to explain much of the difference between hospitals.
The CMS sources contain unavailable values for some hospital-measure combinations.
These values were converted to NULL during the BigQuery cleaning process and were not imputed.
As a result:
some dashboard cells are blank;
the number of hospitals included in each regression changes depending on the selected variables; and
comparisons involving CABG and other less consistently available measures can contain substantially fewer hospitals than broader HCAHPS measures.
A blank dashboard cell should therefore be interpreted as unavailable source data, not a score of zero.
The Tableau story is designed to show three related but distinct dimensions of hospital performance:
Patient Experience: Communication, cleanliness, quietness, discharge information, patient ratings, and willingness to recommend.
Overall CMS Quality: CMS Overall Star Rating.
Clinical Utilization Outcome: Condition-specific 30-day readmission rates.
The strongest relationships occur within the patient-experience measures themselves.
The relationship between those measures and CMS Overall Star Rating is weaker.
The relationship between patient experience and 30-day readmission rates is weaker still.
That distinction is the main conclusion I want the dashboard to communicate: a complete view of hospital performance requires looking at patient experience and clinical performance together rather than assuming one can stand in for the other.