Are there any medical or clinical data sources that have de-identified patient data (name, date of birth, sex) as well as basic physical parameters such as weight, or height? I'm specifically interested in evaluating the medical Entity-Attribute-Value model and for this, various patient parameters would be useful.
-
This article describes some of the issues I'm after.– KermitCommented May 8, 2013 at 19:37
-
I assume that the VA data sets they describe aren't available to you? Obviously I'm sure they're not publicly available, but everyone has access to different data, so I thought I would ask.– John BensinCommented May 8, 2013 at 20:25
-
1Is there a specific medical EAV that you are referring to? Are you looking at NOSQL databases (i.e., key-value pairs) to store the information? What evaluation criteria did you have in mind? I don't see how having the data would help in evaluating a data model.– David MedinetsCommented May 9, 2013 at 1:37
-
if you're looking for survey data of the us population, try brfss, meps, nhanes, or nhis– Anthony DamicoCommented May 30, 2013 at 9:41
-
Thanks for your great answers, everybody! Unfortunately, I do not have enough reputation to comment on the appropriate posts. So I will add a question here. I have a even more specific request. I there any open data of the discussed kind that contain patients' blood values/laboratory data (possibly along with a diagnosis)? thanks!– neuro timCommented Aug 7, 2016 at 12:03
6 Answers
CDC's National Health Interview Survey (HIS) provides public use files containing person-level data including demographic data and physical parameters such as weight, and disability. Dating back to 1957, it is an annual survey of a nationally representative sample dates.
CDC's National Health And Nutritional Examination Survey (NHANES) provides public use files containing person-level data including demographic, dietary, examination, and laboratory data.
The Health and Retirement Study (sponsored by the National Institute on Aging, and Social Security Administration) is longitudinal panel cohort study that started in 1992, and covers a representative sample of Americans over the age of 50, as they transition into retirement. Only certain person-level data are public use, because of the sensitive data collected including genetic information, prescription drug use, cognitive function, biomarkers, etc. The "core" survey is supplemented by focused add-on modules.
The 3 surveys above are used for econometric microsimulation models; in case you wanted another way to search for these datasets that would be one way to back into it.
Note, de-identified patient data will not include names.
Concerning EAV models that use large datasets where information in certain categories is relatively sparse, "sparse" suggests that an individual could be re-identified. At least with US government release of data, any such sparse data would be suppressed if an individual could be re-identified. Protection of personally identifiable information (PII) and maintaining patient confidentiality are considered fundamental to ensuring public trust and participation in these surveys.
Disclosure: I work for the Asst Secretary for Planning and Evaluation (ASPE) in the Dept of Health and Human Services (HHS) on the open data initiative. Surveys mentioned above are all HHS surveys, though I have no direct connection to any of them.
Update: In response to your updated question about EAV, some of the datasets below may prove helpful. If you're looking for fairly large datasets where information in certain categories is relatively sparse, the BRFSS data may prove helpful. Otherwise, the other datasets may or may not serve your purpose.
I've used the data from the CDC's Behavioral Risk Factor Surveillance System (BRFSS), which includes information on a wide array of health factors. It isn't patient data per se, but it's certainly medical data. I've also used the National Longitudinal Survey of Youth, which is a panel data set from the social sciences; it contains a lot of data that's useful for economists, but it does include height and weight. You can look through the summary of variables to see if it contains data you might need.
You could also look at a random sample of the Medical Quality Improvement Consortium patient data set. According to that site:
The records have been stripped of any personal identification and include the following dimensions: gender, age, diabetes, hypertension, stroke, heart disease, smoking history, and BMI.
Another place to look is the National Institute of Mental Health, which maintains a list of clinical datasets that might prove helpful.
MIMIC initiative provides a dataset (deceased patients) that has vitals, Diagnoses, Medications and several other domains.
Updated link (2021) https://mimic.physionet.org/gettingstarted/demo/
Original answer links http://physionet.org/mimic2/demo/ http://physionet.org/mimic2/demo/mimic2dead.sql.gz
Another option is synthea (synthetic data) or OHDSI package Eunomia.
1. dbGaP is a database of Phenotypes and Genotypes. For a project that has approval by local ethics committee (in USA - Institutional Review Board), you can request phenotype data (e.g., blood pressure or eye color is a phenotype element) from dbGaP.
Sample search is here: http://www.ncbi.nlm.nih.gov/gap/?term=framingham
2. www.i2b2.org site has demo data for a system called i2b2 that use EAV structure to store patient parameters
There a cancer registry database from US, you get date of diagnosis, type of cancer, treatment given, etc.
You need to fill, sign and fax a form to SEER to get access (application is easy)
most of the medical data sets posted here have some variation of what you're requesting