Observational Research

As part of the Principal Investigator (PI) team you will work with ARC team members to lead the execution of research. The PI is typically instrumental in developing the research question and translating it into an appropriate study design. The PI has many roles in a project including scientific (e.g., formalize hypotheses), project management (e.g., setting goals, timelines), regulatory (e.g., obtaining IRB approval), and dissemination (e.g., manuscript writing, responding to peer review). Due to the large burden in terms of time and effort, it may be desirable to recruit a co-principal investigator (Co-PI) to share the substantial duties of the role.   

Tasks

•  Develop a research question and place it into the context of what is already known about the topic
• Recruit a team of investigators to assist in the many tasks
• Lead in the development of a protocol, including the development of a statistical analysis plan (SAP)
• Obtain appropriate regulatory approval (e.g., IRB, database procurement, MPOG)
• Manage the conduct of the study/analysis
• Lead in manuscript preparation and submission
• Respond to peer review and lead revision efforts

Time Commitment

This role involves a substantial amount of effort over an extended period of time. Depending on the scope of the project, it would not be unusual for a PI to spend five hours each week for several months. Because the PI (or team of PIs) drives the completion of the project, it is essential that the PI have enough time to devote to the role.

What Can I Learn? 

This role is typically not for first-time investigators, but the PI(s) will become proficient in study design, relevant statistical analyses, bias, confounding, reliability, IRB, manuscript writing, and interacting with peer review

Resources

• Thinking about being a PI? Contact Tim Houle (thoule1@mgh.harvard.edu) 

As part of power analysis team, you will work with our statistician to perform a priori or pragmatic power analysis to estimate appropriate sample sizes as a crucial step for planning your study.

TASKS

• Plan and design your study in a priori fashion;
• Choose appropriate methods for power analysis and sample size estimation;
• Search previous literature to estimate proper effect sizes to be used for power analysis;
• Tuning the power analysis and study design based on customized research needs;
• Summarize and write the power analysis section.

TIME COMMITMENT

The time commitment of this role needs to be adjusted according to the nature of the project. Power analysis usually can be finished less than a week.

WHAT CAN I LEARN?

• Study design and power analysis basics;
• Common methods for power analysis;
• How to use power analysis software PASS;
• Power analysis write-up skills.

RESOURCES

Who to Contact with Reliability Questions: Hao Deng, MPH (hdeng1@mgh.harvard.edu)

Where to Find out More About Power / Sample Size: 

• PASS software for power analysis

As part of data wrangling team, you will work on the fundamental and essential part of data analysis. Data wrangling typically involves the process of transforming the raw data into a desired useful and accurate format for better visualization and statistical modeling. It helps to get better insights at the very beginning and accelerate the decision-making process. Vital discoveries made during the data wrangling phase are valuable and sometimes change the direction of the entire project.

TASKS

• Identifying the extent and relationships of source data and integrate information
• Decompose the data into different structured format
• Solve transformation problems, including merging, ordering and aggregation
• Clean the dataset and deal with the missing values
• Enrich the dataset and avoiding selection bias
• Feature engineering
• Validate for consistency, quality and security 

Time Commitment

The time commitment of this role needs to be adjusted according to the nature of the project. Data wrangling can take a substantial amount of time if done properly. This can take anywhere from 30 minutes to several hours a week. The cleaner and more organized your dataset is, the easier this step will be.

What Can I Learn?

• Use a mix of visual tools and technologies to simplify data
• Lend valuable and actionable insights
• Understand benefits/detriments of the most common data manipulations

RESOURCES

Who to Contact with Questions: Hao Deng, MPH (hdeng1@mgh.harvard.edu) or Xiaojun Xu PhD (xxu@mgh.harvard.edu)

Where to Find out More About Data Wrangling: 

• Tutorial vides and R packages from the RWE lab

As part of the reliability team, you will use your clinical knowledge to identify threats to the interpretation of the research. Reliability is the consistency of a measure across repeated occasions under the same conditions. There are many forms of reliability such as internal consistency, inter-rater agreement, and test-retest. Good observational research studies will consider reliability and measurement error as a core part of the analysis process.

TASKS

• Conduct literature searches for established reliability
• Review records
• Calculate reliability
• Identify sources of unreliability
• Identify consequences of unreliable measurement
• Conduct sensitivity analyses based on measurement error assumptions

TIME COMMITMENT

The time commitment associated with this team will be dependent on the nature of the project. Some projects will involve very simple sensitivity calculations. Other projects will involve records review. The amount of time is expected to range from 1 hour total to several hours per week.

WHAT CAN I LEARN?

• Different types of reliability
• How reliability is assessed
• Sources of measurement error
• How unreliability can affect results
• How to identify unreported measurement error in published studies

RESOURCES

Who to Contact with Reliability Questions: Dana Turner, MSPH PhD (dpturner@mgh.harvard.edu)

Where to Find out More About Reliability: 

• Turner DP, Houle TT. Assessing and interpreting reliability. Headache. 2019;59:653-658. 31032914
• Reliability: https://en.wikipedia.org/wiki/Reliability_%28statistics%29
• Observational error: https://en.wikipedia.org/wiki/Observational_error

As part of the data analysis team you will work with our experienced analysts and statisticians to explore, analyze and interpret clinical data using quantitative analytic methods. Data analysis typically involves the process of transforming or engineering features (e.g. demographics, clinical data, biomedical variables, etc.) of structured data into appropriate forms and analyzing them using complex statistical modelings or machine learning methods.

TASKS

• Variable transformation and feature engineering for analysis
• Defining the hypothesis testing / prediction framework for analysis
• Application of appropriate statistical methods or data science techniques for inferences
• Evaluate the performance, accuracy and generalizability of generated models
• Causal inferences and results interpretation
• Advanced data visualization

TIME COMMITMENT

The time commitment associated with this team depends on the nature of the project. Proper data analysis can represent a significant amount of work and time. This can range from something like 1 hour per week for small projects (e.g. running a chi-square test to compare two proportions) to many hours a week.

WHAT CAN I LEARN?

• Use a mix of visual or programming tools, softwares and packages to analyze data
• Predictive analytics and machine learning methods
• Epidemiology and statistical methods
• Appropriate data interpretation and casual inferences

RESOURCES

Who to Contact with Reliability Questions: Timothy Houle, PhD (thoule1@mgh.harvard.edu) or Hao Deng, MPH (hdeng1@mgh.harvard.edu)   

Where to Find out More About Data Analysis: 

• Tutorial vides and R packages from the RWE lab
• Easy to use Rstudio plugins