rtemis/differentiated-thyroid-cancer-recurrence

Differentiated Thyroid Cancer Recurrence

Clinicopathologic features of 383 patients with well-differentiated thyroid cancer and whether the cancer recurred during follow-up. Each patient was followed for at least 10 years. The dataset is UCI Machine Learning Repository dataset 915 in a single Parquet file with one row per patient.

Structure

data/differentiated-thyroid-cancer-recurrence.parquet: 383 rows, 17 columns. There are no missing values.

Column	Type	Description	Levels, in order
age	integer	Age at diagnosis, in years (15 to 82).
gender	categorical	Gender.	F, M
smoking	categorical	Current smoking.	No, Yes
hx_smoking	categorical	History of smoking.	No, Yes
hx_radiotherapy	categorical	History of radiotherapy.	No, Yes
thyroid_function	categorical	Thyroid function status.	Euthyroid, Clinical Hyperthyroidism, Subclinical Hyperthyroidism, Clinical Hypothyroidism, Subclinical Hypothyroidism
physical_examination	categorical	Finding on physical examination of the thyroid.	Normal, Single nodular goiter-left, Single nodular goiter-right, Multinodular goiter, Diffuse goiter
adenopathy	categorical	Lymphadenopathy on physical examination, by location.	No, Right, Left, Bilateral, Posterior, Extensive
pathology	categorical	Histopathologic type.	Papillary, Micropapillary, Follicular, Hurthle cell
focality	categorical	Tumor focality.	Uni-Focal, Multi-Focal
risk	categorical	American Thyroid Association (ATA) risk of recurrence.	Low, Intermediate, High
t_category	categorical	Primary tumor (T) category of the TNM classification.	T1a, T1b, T2, T3a, T3b, T4a, T4b
n_category	categorical	Regional lymph node (N) category.	N0, N1a, N1b
m_category	categorical	Distant metastasis (M) category.	M0, M1
stage	categorical	Overall stage.	I, II, III, IVA, IVB
response	categorical	ATA response to initial treatment.	Excellent, Indeterminate, Biochemical Incomplete, Structural Incomplete
recurred	categorical	Recurrence during follow-up.	No, Yes

recurred	Patients
No	275
Yes	108

recurred is the last column. Yes, its second level, is the positive class. The dataset has one train split that holds all 383 patients.

Load

R - using rtemis (1.4.3 or later):

dat <- read("hf://datasets/rtemis/differentiated-thyroid-cancer-recurrence")

R - using croissant, with categorical columns as factors in their declared level order:

library(croissant)
dat <- read_record_set(
  "hf://datasets/rtemis/differentiated-thyroid-cancer-recurrence/croissant.json",
  "patients"
)

R - using nanoparquet:

url <- "https://huggingface.co/datasets/rtemis/differentiated-thyroid-cancer-recurrence/resolve/main/data/differentiated-thyroid-cancer-recurrence.parquet"
path <- file.path(tempdir(), "differentiated-thyroid-cancer-recurrence.parquet")
download.file(url, path, mode = "wb")
dat <- nanoparquet::read_parquet(path)

Python - using polars:

import polars as pl
dat = pl.read_parquet("hf://datasets/rtemis/differentiated-thyroid-cancer-recurrence/data/differentiated-thyroid-cancer-recurrence.parquet")

Notes

• response is assessed after initial treatment. A model that uses it predicts recurrence from information that is not available at diagnosis.
• risk and stage are derived from other features, including the TNM categories and age.
• 19 rows have the same values as an earlier row. The dataset has no patient identifier.
• Several levels are rare: hx_radiotherapy Yes (7), adenopathy Posterior (2), stage III (4), and stage IVA (3).
• 312 patients (81.5%) are female.

Source and processing

The UCI file Thyroid_Diff.csv was read and its columns were renamed to snake case: Hx Radiothreapy was named hx_radiotherapy, and T, N, and M were named t_category, n_category, and m_category. Every column except age was stored as a categorical variable with levels in clinical order. The pathology label Hurthel cell was spelled Hurthle cell. No rows were removed.

Metadata

croissant.json is the dataset's Croissant 1.1 metadata, written with the croissant R package. It describes the file with its SHA-256 checksum, each field, and the Responsible AI properties of the data. Each categorical column references an enumeration record set (patients_gender_levels, ..., patients_recurred_levels) that lists its levels in order. The Parquet file stores categorical values as strings.

License

CC BY 4.0, the license of the UCI source dataset.

Citation

Cite the UCI dataset and the paper that describes the cohort:

@misc{borzooei2023differentiated,
  author = {Borzooei, Shiva and Tarokhian, Aidin},
  title = {{Differentiated Thyroid Cancer Recurrence}},
  year = {2023},
  howpublished = {UCI Machine Learning Repository},
  doi = {10.24432/C5632J}
}

@article{borzooei2023machine,
  author = {Borzooei, Shiva and Briganti, Giovanni and Golparian, Mitra and Lechien, Jerome R. and Tarokhian, Aidin},
  title = {Machine learning for risk stratification of thyroid cancer patients: a 15-year cohort study},
  journal = {European Archives of Oto-Rhino-Laryngology},
  year = {2023},
  volume = {281},
  number = {4},
  pages = {2095--2104},
  doi = {10.1007/s00405-023-08299-w}
}