The National Health Ambulatory Medical Care Survey (NHMACS) is an annual national dataset representative of ED visits and resulting hospitalization. Each year the National Center for Health Statistics develops the NHAMCS data collection form. Trained staff then abstract data from patient charts. Using a multi-stage probability sampling design, the dataset captures a representative sample of ALL visits to non-federal hospitals in the United States (US).
Hypertension is independently identifed as a risk factor for cardiovascular disease. Furthermore, acute hypertension can cause immediate life threatening end organ damage, such as, pulmonary edema, cardiac ischemia, intracranial hemorrhage, posterior reversible encephalopathy syndrome, or acute kidney injury. This is referred to as hypertensive emergency. Alternatively, chronic, poorly controlled hypertension can result is significant blood pressure elevation without end organ injury; e.g. asymptomatic hypertension.
The goal of this analysis was to evaluate the charcteristics of patients presenting to United States hospitals and to evaluate specific outcomes for patients with this presentation.
For each observation in the NHAMCS dataset a triage blood pressure is recorded. We used this to evaluate characeristics and outcomes for patients with hypertension above a specified value.
Variables of interest included but are not limited to the following:
- Age, sex, race, insurance status, arrival mode, antihypertensive administration, and use of diagnostic tests (labs and imaging).
- Risk of combined cardiovascular outcomes (stroke, MI, hypertensive emergency), mortality, and admission to the hospitals
For convenience, all the statistical and plotting ouputs are reserved in the repository and can be found in the outputs.
Furthermore, the modified dataset saved as pickle files in the ouptuts folder.
There is a notebook in the 'notebooks' folder that can be used as is to load and review the dataset.
If you want to re-run the analysis or make your own modified analysis then consider the following steps.
git clone https://github.com/coup321/NHAMCS_hypertension
# make sure virtualenv is installed in your base python environment
pip install virtualenv
# create new venv named nhamcs
virtualenv nhamcs
# activate the nhamcs venv
# Windows
.\nhamcs\Scripts\activate
# Mac/Linux
source nhamcs/bin/activate
pip install -r requirements.txt
By default, you can run the package and it will use the dataset in the folder './outputs/working_dataset.pkl' and create all of the tables and plots in the outputs folder.
python NHAMCS_hypertension
If you would like to download the data from the CDC FTP servers then run the package with the force flag. This behavior is also created if the working_dataset.pkl file is somehow missing.
python NHAMCS_hypertension force
Keep in mind that it can take 3-5 minutes to download, unzip, and pickle the files. Furthermore, building the working dataset from the raw data involves multiple large boolean filters over 80,000 x 30 dataframes, so it can take up to 10 minutes to run. This is primarily for medication and 'reason for visit' regular expression filtering.
- RACEETH changes to RACERETH in 2008 and there are some associated coding changes
- IMMED (ESI level) changes to IMMEDR in 2008 and there are some associated coding changes
- PAYTYPE changes to PAYTYPER in 2008 and there are some associated coding changes
Medications are stored in MED1 through MED30 columns. For each medication 1-30 there are associated columnss with info about them.
- GPMED1-30 - indicates if the medication was given, prescribed, or both
- Medication category info as defined in documentation (where # is 1-30)
- RX#CAT1',
- RX#CAT2',
- RX#CAT3',
- RX#CAT4',
- RX#V1C1',
- RX#V1C2',
- RX#V1C3',
- RX#V1C4',
- RX#V2C1',
- RX#V2C2',
- RX#V2C3',
- RX#V2C4',
- RX#V3C1',
- RX#V3C2',
- RX#V3C3',
- RX#V3C4',
What worked for me is to create a boolean filter over the medication info I was interested in, then use that to filter the medications (1-30). If I was just looking for a row that indicated medication was given, for example, then I would reduce the Nx30 filter to 1x30 with an 'any' method on the column axis.