healthcare cms hospitals 2025

Facility ID is a fixed-width 6-character single-token code, unique across all 5421 rows with zero nulls or duplicates. Every value is one word in all caps with length exactly 6, and the sampled tokens (e.g., 010001, 010005) are zero-padded numeric strings consistent with a primary facility identifier. Treatment: Treat as primary key; left-join on this id and exclude from modelling features. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

5,421

len_min

6

len_max

6

len_mean

6

len_median

6

len_p95

6

word_mean

1

word_median

1

n_empty

0

n_duplicates

0

duplicate_rate

0

vocab_size

5,421

readability_flesch_mean

121.2

emoji_rate

0

url_rate

0

one_word_rate

1

allcaps_rate

1

boilerplate_rate

0

This column holds names of healthcare facilities, dominated by terms like 'hospital' (2740), 'center' (1579), and 'medical' (1444), with a typical length of 4 words / 28 characters. Values are near-unique (5286 distinct out of 5421) yet 135 duplicates remain, suggesting either multi-site chains or repeated reporting rows for the same facility. Nearly everything is uppercase (99.3% allcaps rate), which will trip case-sensitive joins. Treatment: Normalize case and whitespace before using as a join key on facility. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

5,286

len_min

3

len_max

74

len_mean

29.21

len_median

28

len_p95

45

word_mean

3.995

word_median

4

n_empty

0

n_duplicates

135

duplicate_rate

0.0249

vocab_size

3,942

readability_flesch_mean

6.842

emoji_rate

0

url_rate

0

one_word_rate

0.001845

allcaps_rate

0.9932

boilerplate_rate

0

Free-text street addresses, with 5387 unique values across 5421 rows (near-unique) and no nulls. 99.2% are all-caps and the top tokens are 'street', 'st', 'avenue', 'drive', 'road', confirming postal-style entries averaging 3.75 words and 19 characters. 34 duplicates exist but no boilerplate, URLs, or emoji. Treatment: Drop or geocode/parse into structured components before modelling; do not feed raw. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

5,387

len_min

7

len_max

50

len_mean

19.37

len_median

19

len_p95

29

word_mean

3.754

word_median

4

n_empty

0

n_duplicates

34

duplicate_rate

0.006272

vocab_size

4,996

readability_flesch_mean

79.27

emoji_rate

0

url_rate

0

one_word_rate

0

allcaps_rate

0.9921

boilerplate_rate

0

This is a US city/town name field, almost entirely uppercased (allcaps_rate 0.994) and dominated by single-word entries (one_word_rate 0.771, word_mean 1.24). Of 5421 rows there are 3049 unique values with a 0.438 duplicate rate, led by CHICAGO (34), HOUSTON (31), and COLUMBUS (23). Lengths are short and tight (len_mean 8.6, len_max 24) and there are no nulls or empties. Treatment: Normalize case and standardize against a city gazetteer (ideally with state) before grouping or joining. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

3,049

len_min

3

len_max

24

len_mean

8.611

len_median

8

len_p95

13

word_mean

1.241

word_median

1

n_empty

0

n_duplicates

2,372

duplicate_rate

0.4376

vocab_size

2,890

readability_flesch_mean

18.29

emoji_rate

0

url_rate

0

one_word_rate

0.7709

allcaps_rate

0.9943

boilerplate_rate

0

US state codes covering 56 distinct values across 5,421 rows with no nulls — likely all 50 states plus territories/DC. Distribution is broad and near-uniform (entropy ratio 0.917), with TX leading at only 8.5% and CA, FL, IL, OH following. The 56 cardinality is worth noting since it exceeds the standard 50 states. Treatment: one-hot or target-encode for modelling; optionally group rare territories. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

56

top_value

TX

top_rate

0.08522

cardinality

56

entropy

5.328

entropy_ratio

0.9174

This is a US ZIP code field stored as a numeric column, with 4721 unique values across 5421 rows and no nulls. The range (603 to 99929) and broad IQR (32771-76104) are consistent with national ZIP coverage, and leading-zero ZIPs (e.g., the min of 603) have already been corrupted by numeric storage. Treating the mean (53780) or std (27064) as meaningful is misleading since ZIPs are categorical identifiers, not quantities. Treatment: Cast to zero-padded 5-character strings and treat as a categorical/geographic key rather than a numeric feature. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

4,721

min

603

max

99,929

mean

5.378e+04

median

55,066

std

2.706e+04

q1

32,771

q3

76,104

iqr

43,333

skew

-0.1646

kurtosis

-0.9879

n_outliers

0

outlier_rate

0

zero_rate

0

This column holds U.S. county or parish names, stored uppercase and almost always as a single token (one_word_rate 0.87, allcaps_rate 1.0). With 1,555 uniques across 5,421 rows and a 71.3% duplicate rate, common counties repeat heavily — LOS ANGELES (88), JEFFERSON and COOK (59 each) lead. Note that vocab_size (1591) exceeds n_unique (1555), implying multi-word names like SAN/LOS/ST. variants contribute extra tokens. Treatment: Normalize casing and join to a state column before using as a categorical geographic feature. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

1,555

len_min

3

len_max

25

len_mean

7.34

len_median

7

len_p95

11

word_mean

1.135

word_median

1

n_empty

0

n_duplicates

3,866

duplicate_rate

0.7132

vocab_size

1,591

readability_flesch_mean

34.44

emoji_rate

0

url_rate

0

one_word_rate

0.8733

allcaps_rate

1

boilerplate_rate

0

This column holds US-style telephone numbers, every value exactly 14 characters long with a mean of 2 words, consistent with a `(NPA) NNN-NNNN` format. Of 5421 rows, 5383 are unique and 38 duplicates appear (0.7%), with no nulls; the most common tokens are area codes like (406), (605), and (402). The near-unique cardinality and rigid length make this an identifier rather than a feature. Treatment: Drop for modelling; retain as a contact identifier or parse out the area code if geography is useful. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

5,383

len_min

14

len_max

14

len_mean

14

len_median

14

len_p95

14

word_mean

2

word_median

2

n_empty

0

n_duplicates

38

duplicate_rate

0.00701

vocab_size

5,550

readability_flesch_mean

120.2

emoji_rate

0

url_rate

0

one_word_rate

0

allcaps_rate

1

boilerplate_rate

0

Categorical classifier of hospital facilities across 8 types, with no nulls in 5421 rows. Acute Care Hospitals dominate at 57.6% (3120 records), followed by Critical Access (1375) and Psychiatric (626); the long tail is sparse, with only 4 Long-term facilities. Entropy ratio of 0.55 confirms the distribution is heavily concentrated in the top category. Treatment: One-hot encode, optionally collapsing the four rarest types into an 'Other' bucket. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

8

top_value

Acute Care Hospitals

top_rate

0.5755

cardinality

8

entropy

1.654

entropy_ratio

0.5513

Categorical descriptor of hospital ownership type across 5421 records, with 12 distinct categories and no nulls. Distribution is dominated by 'Voluntary non-profit - Private' at 42.26% of rows, followed by 'Proprietary' at 1067 and a long tail down to 'Government - Federal' at 44. Entropy ratio of 0.72 indicates moderate concentration but reasonable spread across the taxonomy. Treatment: One-hot or target-encode for modelling; consider collapsing rare tiers like 'Physician' and 'Government - Federal'. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

12

top_value

Voluntary non-profit - Private

top_rate

0.4226

cardinality

12

entropy

2.586

entropy_ratio

0.7215

Binary Yes/No flag indicating whether emergency services were involved or available, with no nulls across 5421 rows. The distribution is imbalanced: 'Yes' accounts for 83.1% (4505) versus 916 'No' responses, giving an entropy ratio of 0.655. Treatment: Encode as a binary 0/1 indicator; be mindful of class imbalance if used as a target. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

2

top_value

Yes

top_rate

0.831

cardinality

2

entropy

0.6553

entropy_ratio

0.6553

This appears to be a binary flag indicating whether a hospital meets criteria for a 'birthing friendly' designation, but it functions as a presence indicator only. Of 5421 rows, 58.24% are null and the remaining 2264 are all 'Y' — there are no 'N' values, giving cardinality 1 and zero entropy. Nulls effectively mean 'not designated' rather than missing data. Treatment: Recode nulls as 'N' to form a usable binary indicator, or drop since it carries the same information as the null mask. high · anthropic:claude-opus-4-7

n

5,421

nulls

3,157 (58.2%)

unique

1

top_value

Y

top_rate

1

cardinality

1

entropy

0

entropy_ratio

0

This is the CMS-style 1-5 hospital star rating stored as a string, with a sixth bucket 'Not Available' for unrated facilities. The striking issue is that 'Not Available' dominates at 47.1% of 5,421 rows, outnumbering every actual rating; among rated hospitals, 3 stars (937) is most common and 1 star (229) least. Entropy ratio of 0.83 across 6 categories confirms the distribution is spread but heavily anchored by missingness-as-category. Treatment: Recode 'Not Available' to NaN and treat the remainder as an ordinal 1-5 scale. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

6

top_value

Not Available

top_rate

0.4708

cardinality

6

entropy

2.133

entropy_ratio

0.8252

Footnote codes attached to the Hospital overall rating, with only 7 distinct values across 5,421 rows. Over half the column is null (null_rate 0.527), and among populated rows code '16' dominates at 65.4% followed by '19', so the field carries low information (entropy_ratio 0.41). One compound entry '16, 23' appears twice, indicating values can be multi-coded. Treatment: Keep as a categorical flag alongside the rating; split the rare compound codes if you need per-footnote analysis. high · anthropic:claude-opus-4-7

n

5,421

nulls

2,857 (52.7%)

unique

7

top_value

16

top_rate

0.6537

cardinality

7

entropy

1.158

entropy_ratio

0.4126

This is a binary categorical field indicating the count of measures in a mortality (MORT) group, holding only the value '7' (84.1% of rows) or 'Not Available' (the remaining 863 rows). Despite being labeled a count, it functions as a presence flag: when data exists, the count is always 7. No nulls are recorded, since missingness is encoded as the literal string 'Not Available'. Treatment: Recode to a boolean availability flag ('7' vs 'Not Available') before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

2

top_value

7

top_rate

0.8408

cardinality

2

entropy

0.6324

entropy_ratio

0.6324

Likely the count of facility mortality (MORT) measures reported per hospital, stored as strings 1-7 alongside a 'Not Available' sentinel. The dominant value is 'Not Available' at 32.8% (1777/5421), which makes the column nominally categorical with 8 levels even though 7 of them are integers. Entropy ratio of 0.92 indicates the non-null counts are spread fairly evenly across 1-7. Treatment: Replace 'Not Available' with NaN, cast remaining values to integer, and optionally add a missingness indicator. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

8

top_value

Not Available

top_rate

0.3278

cardinality

8

entropy

2.765

entropy_ratio

0.9217

Categorical count of mortality measures on which a hospital performed better than the national rate, encoded as small integers stored as strings (0–7) plus a 'Not Available' sentinel. The distribution is heavily concentrated at 0 (57.8% of 5421 rows) with another 1777 rows marked 'Not Available', leaving only ~10% of hospitals showing any 'better' measure. Entropy ratio of 0.458 confirms the skew, and the sentinel string blocks direct numeric use. Treatment: Replace 'Not Available' with NaN, cast remaining values to int, and consider binarising (any-better vs none) given the skew. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

9

top_value

0

top_rate

0.5785

cardinality

9

entropy

1.453

entropy_ratio

0.4583

This column counts how many mortality (MORT) measures a hospital scored 'No Different' on, encoded as a small integer 0-7 but stored as strings alongside a 'Not Available' sentinel. Cardinality is just 9 with high entropy ratio (0.885), meaning values are spread fairly evenly across the integers — except 'Not Available' dominates at 32.8% (1777/5421) and '0' is rare at only 12 rows. Analysts should note the heavy missing-as-string encoding and the near-empty '0' bucket. Treatment: Recode 'Not Available' to null and cast remaining values to integer before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

9

top_value

Not Available

top_rate

0.3278

cardinality

9

entropy

2.806

entropy_ratio

0.8852

This column counts how many mortality (MORT) measures a hospital scored worse than the national average, encoded as small integers 0-5 with a 'Not Available' sentinel mixed in. Roughly 60% of 5421 rows are '0' and another 1777 are 'Not Available', leaving only 378 hospitals flagged on any MORT measure and just 1 hospital at the maximum of 5. The mixed string/integer encoding is the main gotcha. Treatment: Replace 'Not Available' with NaN and cast to integer before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

7

top_value

0

top_rate

0.6025

cardinality

7

entropy

1.294

entropy_ratio

0.4608

This appears to be a footnote code attached to MORT (mortality) group records, stored as a small numeric category with only 4 distinct values ranging from 5 to 23. Two-thirds of rows are null (null_rate 0.672), suggesting the footnote applies only to a minority of records. The strongly negative kurtosis (-1.96) and bimodal-looking quartiles (Q1=5, Q3=19) confirm these are discrete codes rather than a continuous measurement. Treatment: Cast to categorical and treat nulls as 'no footnote' rather than imputing numerically. high · anthropic:claude-opus-4-7

n

5,421

nulls

3,643 (67.2%)

unique

4

min

5

max

23

mean

11.58

median

5

std

7.057

q1

5

q3

19

iqr

14

skew

0.1488

kurtosis

-1.959

n_outliers

0

outlier_rate

0

zero_rate

0

Binary categorical with only two values across 5421 rows: the literal string "8" (84.1%) and "Not Available" (15.9%). Despite the name suggesting a count, it functions as a flag indicating either a fixed measure count of 8 or missing data encoded as text. Treatment: Recode "Not Available" to null and binarize, or drop given near-constant value. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

2

top_value

8

top_rate

0.8408

cardinality

2

entropy

0.6324

entropy_ratio

0.6324

This column reports the count of facility safety measures, stored as a categorical with 9 distinct values (1-8 plus 'Not Available'). The dominant value is 'Not Available' at 38.1% (2,065 of 5,421 rows), making missingness encoded as a string rather than null. Among reported counts, 7 is most common (733), and the distribution across 1-8 is uneven rather than monotonic. Treatment: Recode 'Not Available' to null and cast remaining values to integer before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

9

top_value

Not Available

top_rate

0.3809

cardinality

9

entropy

2.753

entropy_ratio

0.8684

This is a small-integer count (0-6) of safety measures rated 'Better', stored as strings alongside a 'Not Available' sentinel. The sentinel dominates at 38.1% (2065/5421) of rows, and the distribution is heavily right-skewed: 0 and 1 cover 2600 rows while 5 and 6 appear only 14 and 3 times. Cardinality is 8 with entropy ratio 0.70, so most signal sits in the lower bins. Treatment: Recode 'Not Available' to missing, cast remaining values to integer, and consider binning the sparse 4-6 tail. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

8

top_value

Not Available

top_rate

0.3809

cardinality

8

entropy

2.11

entropy_ratio

0.7033

A small-integer count (0-8) of safety measures rated 'no different', stored as strings alongside a 'Not Available' sentinel. The sentinel dominates at 38.1% of 5421 rows (2065), making it the modal value ahead of any actual count. Among real values, 5, 2, 4, and 1 cluster between 509-656 occurrences while 0 and 8 are rare (20 and 10). Treatment: Cast numeric levels to int and treat 'Not Available' as an explicit missing indicator before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

10

top_value

Not Available

top_rate

0.3809

cardinality

10

entropy

2.685

entropy_ratio

0.8083

This is a low-cardinality count column (5 distinct values) recording how many safety measures got worse, taking integer values 0-3 plus a 'Not Available' sentinel. Most rows (54.3%) are 0 and another 2065 are 'Not Available', leaving only 415 rows with any worsening measures (1-3). The mix of numeric strings and a textual missing-marker means it isn't cleanly numeric as stored. Treatment: Recode 'Not Available' to NaN and cast to integer before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

5

top_value

0

top_rate

0.5425

cardinality

5

entropy

1.338

entropy_ratio

0.5764

This appears to be a footnote code column tied to a 'Safety Group' classification, encoded numerically but with only 4 distinct values ranging from 5 to 23. 61.8% of rows are null, suggesting footnotes apply only to a minority of records. The bimodal-looking distribution (median 5, q3 19, kurtosis -1.81) indicates these are categorical flags rather than a true measurement. Treatment: Cast to categorical and treat nulls as 'no footnote' rather than imputing numerically. high · anthropic:claude-opus-4-7

n

5,421

nulls

3,350 (61.8%)

unique

4

min

5

max

23

mean

10.69

median

5

std

6.95

q1

5

q3

19

iqr

14

skew

0.4116

kurtosis

-1.809

n_outliers

0

outlier_rate

0

zero_rate

0

A categorical column with only two distinct values: the count '11' covers 84.1% of 5421 rows, with the remaining 863 rows marked 'Not Available'. Functionally this is a binary availability flag rather than a true count, since '11' is the only numeric value present. The mixing of a numeric string with a sentinel 'Not Available' is the main surprise. Treatment: Recode to a binary available/not-available indicator before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

2

top_value

11

top_rate

0.8408

cardinality

2

entropy

0.6324

entropy_ratio

0.6324

This column reports the count of facility readmission (READM) measures available per facility, stored as strings rather than integers. With 12 distinct values and entropy ratio 0.965, the values are spread fairly evenly across small integers, but the modal value is the literal string "Not Available" at 21.2% (1150 of 5421 rows), which masks missingness as a category. Numeric values like "11", "8", "6", and "9" dominate the rest. Treatment: Replace "Not Available" with null and cast to integer before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

12

top_value

Not Available

top_rate

0.2121

cardinality

12

entropy

3.459

entropy_ratio

0.965

Counts how many readmission measures a hospital scored 'better than national rate' on, encoded as small integers 0-5 with 'Not Available' as a seventh category. The distribution is heavily concentrated at 0 (61.5% of 5421 rows) and ties off rapidly: only 28 rows hit 3, 10 hit 4, and 3 hit 5. Notably, 'Not Available' (1150 rows) is the second most common value, larger than every nonzero count combined. Treatment: Cast numeric levels to int and treat 'Not Available' as an explicit missing-category flag before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

7

top_value

0

top_rate

0.6146

cardinality

7

entropy

1.51

entropy_ratio

0.5379

This column counts how many readmission (READM) measures at a hospital scored 'No Different' than the national rate, encoded as integers 1-9+ alongside a 'Not Available' sentinel. With 13 unique values and entropy ratio 0.92, the distribution is fairly flat across counts 1-9 (each ~370-500 rows), but 'Not Available' is the single largest bucket at 21.2% (1,150 of 5,421). That missingness-as-string is the main surprise and would corrupt any numeric aggregation if not handled. Treatment: Cast to integer with 'Not Available' converted to NaN before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

13

top_value

Not Available

top_rate

0.2121

cardinality

13

entropy

3.408

entropy_ratio

0.9211

This appears to be a hospital-level count of readmission measures performing worse than the national rate, stored as a categorical/string field with values 0-7 plus a 'Not Available' sentinel. The distribution is heavily concentrated at '0' (55.1% of 5421 rows) with another 1150 rows marked 'Not Available', leaving genuine non-zero counts as a long tail that thins to just 1-2 hospitals at values 5, 6, and 7. The mixing of numeric strings with 'Not Available' is the main gotcha. Treatment: Replace 'Not Available' with NaN and cast remaining values to integer before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

9

top_value

0

top_rate

0.5512

cardinality

9

entropy

1.758

entropy_ratio

0.5545

Likely a footnote/flag code attached to a hospital readmission group metric, encoded as a small integer rather than a true numeric quantity. Only 3 distinct values appear (min 5, max 22, median 19) and 78.79% of rows are null, so the column is sparse and categorical in nature. The non-null distribution is concentrated at the higher codes, with q1=5 and q3=19 indicating a bimodal split between two code clusters. Treatment: Cast to categorical footnote code and treat nulls as 'no footnote' rather than imputing numerically. high · anthropic:claude-opus-4-7

n

5,421

nulls

4,271 (78.8%)

unique

3

min

5

max

22

mean

15.15

median

19

std

6.366

q1

5

q3

19

iqr

14

skew

-0.9528

kurtosis

-1.051

n_outliers

0

outlier_rate

0

zero_rate

0

A binary categorical with only two values: the string "8" covering 84.08% of 5,421 rows and "Not Available" for the remaining 863. This looks like a measure-count field that is fixed at 8 when reported, otherwise marked as missing via a sentinel rather than a true null (null_rate is 0.0). Entropy ratio of 0.63 reflects that imbalance. Treatment: Convert "Not Available" to a true null and treat as a binary missingness indicator; the column carries little signal otherwise. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

2

top_value

8

top_rate

0.8408

cardinality

2

entropy

0.6324

entropy_ratio

0.6324

This column is a binary categorical indicator with only two distinct values across all 5,421 rows: the literal string "8" (58.2%) and "Not Available" (41.8%). Despite being labeled as a count, it never varies numerically — every reporting facility either has exactly 8 measures or none reported, making this effectively a presence/absence flag with high entropy ratio (0.98). The 41.8% "Not Available" rate is a substantial missingness signal masquerading as a category. Treatment: Recode as a binary available/not-available flag rather than treating as a numeric count. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

2

top_value

8

top_rate

0.5818

cardinality

2

entropy

0.9806

entropy_ratio

0.9806

This appears to be a footnote code attached to a 'Pt Exp Group' (patient experience group) measure, stored as a numeric but functioning as a categorical flag. Only 3 distinct values appear (min 5, median 5, max 22) and 58.18% of rows are null, indicating footnotes are applied sparingly to a minority of records. The bimodal-looking spread (Q1=5, Q3=19, kurtosis -1.66) confirms this is a small set of discrete codes rather than a true numeric measurement. Treatment: Cast to categorical footnote code and treat nulls as 'no footnote' rather than missing. high · anthropic:claude-opus-4-7

n

5,421

nulls

3,154 (58.2%)

unique

3

min

5

max

22

mean

10.15

median

5

std

6.806

q1

5

q3

19

iqr

14

skew

0.571

kurtosis

-1.658

n_outliers

0

outlier_rate

0

zero_rate

0

Binary categorical column with only two values: '12' (84.1% of rows) and 'Not Available' (the remaining 863 rows). The literal '12' suggests this captures a count of measures per TE group, but it is stored as a string and effectively constant when present. The 'Not Available' sentinel functions as the only real source of variation. Treatment: Convert to a boolean is_available flag; drop the constant '12' value. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

2

top_value

12

top_rate

0.8408

cardinality

2

entropy

0.6324

entropy_ratio

0.6324

This column reports the count of facility TE (timely & effective care) measures available per record, but it's stored as strings mixing numeric values ("4" through "12") with the sentinel "Not Available". The most common value is "Not Available" at 17.1% (928 of 5421), making missingness the modal category despite a reported null_rate of 0. Entropy ratio of 0.93 across 13 categories shows the non-null counts are spread fairly evenly, with "10" and "11" the dominant numeric values. Treatment: Recode "Not Available" to null and cast remaining values to integer before modelling. high · anthropic:claude-opus-4-7

n

5,421

nulls

0 (0.0%)

unique

13

top_value

Not Available

top_rate

0.1712

cardinality

13

entropy

3.458

entropy_ratio

0.9343

This appears to be a footnote reference code for a 'TE Group', stored numerically but functioning as a categorical pointer with only 3 distinct values across 5421 rows. It is overwhelmingly empty (82.88% null) and heavily concentrated at 19 (both q1 and q3 equal 19, iqr=0), producing a strong left skew (-2.43) and 133 outliers (14.33%) at the low end down to 5. The numeric stats like mean=17.58 are misleading given the column is effectively a small code set. Treatment: Cast to categorical and treat nulls as 'no footnote'; do not use as a numeric feature. high · anthropic:claude-opus-4-7

n

5,421

nulls

4,493 (82.9%)

unique

3

min

5

max

22

mean

17.58

median

19

std

4.432

q1

19

q3

19

iqr

0

skew

-2.43

kurtosis

4.12

n_outliers

133

outlier_rate

0.1433

zero_rate

0