Datasets

tausurv.datasets is a registry of survival cohorts that download on first use, verify against a pinned SHA256, and cache content-addressed.

from tausurv.datasets import load_dataset, list_datasets, list_variants

list_datasets()                     # every loadable table
list_datasets(base_only=True)       # one name per study
list_datasets(tag="competing-risks")

X, time, event = load_dataset("pbc")

Names, variants and aliases

A name is base or base:variant. The base name resolves to the reading a survival textbook would call default; a variant names another published processing of the same rows. This matters because the literature does not agree on a single canonical table for most cohorts, and a score is only interpretable against the table it was computed on.

list_variants("pbc")
# ['pbc', 'pbc:randomised', 'pbc:transplant']

Names that refer to different tables in different sources:

you may mean	registry name	what it is
GBSG / GBSG2	gbsg	the 686-patient German trial
GBSG (DeepSurv)	gbsg_rotterdam	2232 rows, Rotterdam + German trial combined
PBC	pbc / pbc:randomised	all 418 patients / the 312 randomised ones
SUPPORT	support / support:nonleaky	raw 47 columns / baseline covariates only

Raw SUPPORT ships the SUPPORT prognostic model’s own survival estimates as covariates: surv6m alone scores c-index 0.72 against the endpoint, above published models fitted on real covariates. support:nonleaky removes those columns. The raw entry stays, so the leak stays inspectable.

Aliases (gbsg2, actg320, diabetic, …) resolve but never appear in list_datasets(), so each table has exactly one canonical name.

Access tiers

Most datasets are OPEN and need no setup. SEER, UNOS, MIMIC, full METABRIC and Framingham cannot be redistributed; their specs and parsers ship anyway, and loading one raises with the steps to obtain it:

load_dataset("seer", path="/path/to/my/seer/export")

Four benchmark tables are distributed as HDF5 and need pip install 'tausurv[deepsurv]': metabric, gbsg_rotterdam, support:deepsurv, whas_deepsurv. Those files carry no column names, so their covariates are exposed positionally as x0, x1, … rather than under names inferred from a paper and impossible to verify against the data.

Catalogue

Each point is one open study in its base reading; variants and gated datasets are omitted. The event rate is the fraction of subjects with an observed event of any cause. Sizes range from 23 subjects (aml) to 17549 (nafld), event rates from 2.1% (pneumon) to 100% (hoel, ifluid).

dataset	n	d	events	outcome	tags
aids	1151	11	8.3%	single	clinical,trial,benchmark,rare-events
aids:death	1151	11	2.3%	single	clinical,trial,rare-events
alloauto	101	1	49.5%	single	clinical,transplant,non-proportional
allograft	34	1	85.3%	single	clinical,transplant,clustered
aml	23	1	78.3%	single	clinical,trial
ashkenazi	3920	1	12.1%	single	clinical,genetics,clustered
baboon	152	0	38.2%	single	ecology,left-truncated
bfeed	927	8	96.2%	single	behavioural,survey
bladder	85	3	55.3%	single	clinical,trial,first-event
bmt	137	11	60.6%	single	clinical,transplant
bmt:competing	137	11	60.6%	2 causes	clinical,transplant,competing-risks
bnct	30	1	90.0%	single	preclinical,trial
breast_cancer_gse7390	198	80	25.8%	single	clinical,genomics,high-dimensional
btrial	45	1	53.3%	single	clinical
burn	154	11	31.2%	single	clinical,trial
capacitor	64	2	50.0%	single	reliability
cgd	128	10	34.4%	single	clinical,trial,first-event
channing	462	2	38.1%	single	clinical,left-truncated
colon	929	12	54.5%	2 causes	clinical,competing-risks
colon:cause1	929	12	50.4%	single	clinical
colon:cause2	929	12	4.1%	single	clinical
colon:death	929	12	48.7%	single	clinical,trial
colon:recurrence	929	12	50.4%	single	clinical,trial
drughiv	34	1	79.4%	single	clinical
flchain	7874	8	27.5%	single	clinical
flchain:complete	6524	8	30.1%	single	clinical
flchain:positive	7871	8	27.5%	single	clinical
framingham	-	-	-	credentialed	clinical,epidemiology,cardiology,longitudinal
gastric_xelox	48	0	66.7%	single	clinical,single-arm
gbsg	686	8	43.6%	single	clinical
gbsg_rotterdam	2232	8	56.8%	single	clinical,benchmark,breast-cancer
genfan	70	0	17.1%	single	reliability
hepatocellular	227	43	42.7%	single	clinical,biomarkers
hepatocellular:rfs	227	43	63.0%	single	clinical,biomarkers
hodgkins	43	4	60.5%	single	clinical,transplant
hoel	181	1	100.0%	3 causes	competing-risks,preclinical
hoel:cause1	181	1	28.2%	single	preclinical
hoel:cause2	181	1	29.3%	single	preclinical
hoel:cause3	181	1	42.5%	single	preclinical
ifluid	41	1	100.0%	single	reliability
imotor	40	1	42.5%	single	reliability
kidney_catheter	76	3	76.3%	single	clinical,frailty,clustered
kidney_infection	119	1	21.8%	single	clinical
kidney_transplant	863	4	16.2%	single	clinical
larynx	90	3	55.6%	single	clinical
leukemia	42	2	71.4%	single	clinical,trial,clustered
lung	228	8	72.4%	single	clinical
lung:complete	168	7	72.0%	single	clinical
melanoma	205	5	34.6%	2 causes	clinical,competing-risks
melanoma:cause1	205	5	27.8%	single	clinical
melanoma:cause2	205	5	6.8%	single	clinical
metabric	1904	10	57.9%	single	clinical,benchmark,breast-cancer
metabric:full	-	-	-	credentialed	clinical,genomics,breast-cancer,high-dimensional
mgus	241	7	94.2%	2 causes	clinical,competing-risks
mgus2	1384	6	70.4%	2 causes	clinical,competing-risks
mgus2:cause1	1384	6	8.3%	single	clinical
mgus2:cause2	1384	6	62.1%	single	clinical
mgus:cause1	241	7	26.6%	single	clinical
mgus:cause2	241	7	67.6%	single	clinical
mgus:death	241	7	93.4%	single	clinical
mimic	-	-	-	credentialed	clinical,critical-care,ehr,large
myeloid	646	3	49.5%	single	clinical,trial
myeloma	3882	2	71.3%	single	clinical,left-truncated
nafld	17549	5	7.8%	single	clinical,epidemiology
nwtco	4028	6	14.2%	single	clinical
ovarian	26	4	46.2%	single	clinical
pbc	418	17	38.5%	single	clinical
pbc:randomised	312	17	40.1%	single	clinical,trial
pbc:transplant	418	17	44.5%	2 causes	clinical,competing-risks
pharmaco_smoking	125	11	71.2%	single	clinical,trial,behavioural
pneumon	3470	13	2.1%	single	clinical,epidemiology,rare-events
prostate	14294	3	28.3%	2 causes	clinical,competing-risks,registry
prostate:cause1	14294	3	5.6%	single	clinical,registry
prostate:cause2	14294	3	22.7%	single	clinical,registry
psych	26	2	53.8%	single	clinical,left-truncated
retinopathy	394	6	39.3%	single	clinical,clustered
rossi	432	7	26.4%	single	recidivism
rotterdam	2982	10	42.7%	single	clinical,breast-cancer
rotterdam:recurrence	2982	10	50.9%	single	clinical,breast-cancer
rotterdam:rfs	2982	10	57.4%	single	clinical,breast-cancer,benchmark
seer	-	-	-	user_provided	clinical,competing-risks,registry,large
stanford2	184	2	61.4%	single	clinical
std	877	21	39.6%	single	clinical,epidemiology
support	9105	45	68.1%	single	clinical
support:deepsurv	8873	15	68.0%	single	clinical,benchmark,critical-care
support:nonleaky	9105	31	68.1%	single	clinical,benchmark
telco_churn	7043	18	26.5%	single	churn
tongue	80	1	66.2%	single	clinical
transplant	815	4	90.7%	3 causes	clinical,competing-risks
transplant:cause1	815	4	78.0%	single	clinical
transplant:cause2	815	4	4.5%	single	clinical
transplant:cause3	815	4	8.1%	single	clinical
twins	24	1	33.3%	single	epidemiology,clustered
udca	170	4	42.4%	single	clinical,trial
unos	-	-	-	user_provided	clinical,transplant,registry,large
valve_seat	41	0	58.5%	single	reliability,first-event
veteran	137	6	93.4%	single	clinical
waltons	163	1	95.7%	single	clinical
whas500	500	14	43.0%	single	clinical,benchmark,cardiology
whas_deepsurv	1638	7	42.1%	single	clinical,benchmark,cardiology