Sign In

Monika Wahi

Epidemiology & Biostatistics Consultant a/k/a Data Scientist | Exclusive and innovative solutions for data science challenges in public health, research and education

Monika Wahi is an epidemiologist, biostatistician, and data scientist based in Boston, Massachusetts.1 She is the President of DethWench Professional Services, a company that provides educational and research consulting services in data science and public health.1

Professional Experience

  • Founded DethWench Professional Services in January 2004, offering expertise in study design, epidemiologic consultation, data analysis, grantwriting, and more.1
  • Served as a Lecturer at Laboure College from 2012 to 2018, teaching healthcare and statistics courses.1
  • Worked as an Epidemiologist for the US Army Research Institute for Environmental Medicine from 2008 to 2011.1
  • Held the position of Epi-Informaticist at Byrd Alzheimer's Institute from 2003 to 2008.1

Education and Certifications

  • Pursued a Ph.D. in Epidemiology at the University of South Florida (2004-2008, ABD).1
  • Holds an MPH in Epidemiology from the University of Minnesota-Twin Cities (2000-2003).1
  • Certified in Public Health (CPH) by the National Board of Public Health Examiners since 2008.1

Additional Roles

Monika is also a LinkedIn Learning author, having published 8 data science courses on the platform.23 She is known for her engaging teaching style and ability to make complex topics accessible.2

Highlights

Sep 2 · linkedin

I read one scientific journal found in 2018 that there were about 180 biological databases.

In addition, there has been an explosion of genome databases for many different organisms.

One of the main challenges with working with genomic data is that they have many intercorrelations.

Dealing with many, highly-correlated variables is a statistical challenge.

I often do regression, and I want to use these variables as independent variables in the regression.

However, putting highly correlated variables into the regression together will introduce collinearity and mess up the model.

How can we reduce these variables?

One way is to do descriptive analysis, and make some choices.

  1. If you find certain variables are intercorrelated, and you understand them, just choose the one for your model that you think has the most impact, and leave the others out.
  2. Another choice you can make is to create a composite index of the correlated variables.
  3. Propensity matching is another approach, but it is time-consuming and you end up having to remove some data.

Factor analysis is another approach you can use.

  1. If you have multiple numeric variables, factor analysis can help you figure out which ones are intercorrelated.
  2. You need to judge which variables load on “factors”.
  3. Once you agree on the patterns, you can combine the variables on the same factor into one variable.

Something that is actually more important than these techniques is designing your research carefully before you approach modeling.

If you already have knowledge about the potential variables you could use, rule ones out that you know you couldn’t explain in the model anyway.

Do you have a strategy for reducing the variables you use in models? Share your advice in the comments!

Master the art of analyzing data originating in applications, and impress your colleagues with unique insights! Join our upcoming online workshop, “Application Basics: Integrating Systems”: https://lnkd.in/eZbfdG9V

Sep 1 · linkedin

As of 2021, 88% of US office-based physicians were using an Electronic Health Record (EHR).

According to a 2019 report, market share for the top three EHR companies in the US were Epic (28%), Cerner (26%), and Meditech (16%).

Many of the variables in an EHR are continuous variables. Examples are systolic blood pressure (SBP), body mass index (BMI), and laboratory values.

Many times in statistics, we need to categorize these continuous variables. Here are my strategies.

It’s best to start by running histograms and box plots to observe the distribution of the continuous variable.

Here are the three main ways I approach categorizing a continuous variable. 🟠Visually 🟠Empirically 🟠Data-driven

To select cutpoints visually:

  1. Run the histogram and boxplot.
  2. Also run summary statistics.
  3. Note where there are natural breaks in the data.
  4. Use these categories.

As an example, I often categorize age using the visual approach.

To set cutpoints empirically, I choose an established system. 🟣For SBP, I could use cutpoints for hypertension levels. 🟣For BMI, I could use cutpoints for obesity levels.

The third way I could choose is data-driven. 🔵If the distribution appears bimodal, then I could create a two-level variable and set the cutpoint at the median. 🔵Tertiles, quartiles, and quintiles are also possible data-driven ways to categorize the variable.

One advantage of categorizing continuous variables is that you can create levels for “unknown”, as well as extreme values.

Although you lose data when you categorize a continuous variable, sometimes the categories let you create models that are easier to interpret.

What advice do you have for categorizing continuous variables? Put your tips in the comments!

Master the art of analyzing data originating in applications, and impress your colleagues with unique insights! Join our upcoming online workshop, “Application Basics: Integrating Systems”: https://lnkd.in/eZbfdG9V

Related Questions

What are some notable projects Monika Wahi has led at DethWench Professional Services?
How has Monika Wahi's experience as an epidemiologist influenced her approach to data science?
What specific skills does Monika Wahi teach in her LinkedIn Learning courses?
Can you provide examples of Monika Wahi's contributions to public health initiatives?
How did Monika Wahi transition from the US Army to her current role at DethWench Professional Services?
Monika Wahi
Monika Wahi, photo 1
Monika Wahi, photo 2
Add to my network

Location

Boston, Massachusetts, United States