Tighe Lab

Projects

Machine Learning & Acute Pain Prediction

Our work centers on the application of machine learning to forecast severe acute postoperative pain.

Given the increasingly complex data contained within modern electronic medical record systems, novel approaches are necessary to efficiently and accurately classify data according to a broad array of outcomes.

In one example, our group demonstrated that machine learning algorithms offer multiple advantages over traditional regression-based approaches to classification in a retrospective cohort of patients undergoing ACL reconstruction. (http://www.ncbi.nlm.nih.gov/PubMed/21899717?dopt=Abstract)

Future efforts will center on broader classification experiments examining a broad array of surgeries. These efforts will also integrate both structured variables as well as unstructured data contained within text documents.


Stochastic Process Modeling of Acute Postoperative Pain

Recent work by Chapman and others has revealed substantial heterogeneity in the rate at which acute postoperative pain resolves.

Our team has employed a variety of techniques including symbolic aggregate approximation, Markov chainsmixed-model regression, and dynamic probabilistic graphical models in order to characterize and predict personalized postoperative pain trajectories.


Text and Sentiment Analysis of Pain-Related Documents

Text analysis encompasses the quantitative evaluation of raw text documents.

Here, we convert the terms, phrases, sentiment, arrangement and organization into features which can be included into machine learning algorithms. Text information can thus be inserted directly into classification schemas to assist with outcome forecasting.

This information can be gleaned at several different levels of interpretation. For instance, we may wish to initially check for certain key terms such as “painful”, “extensive”, or “orthopedic” in identifying those clinical documents associated with severe postoperative pain. We can also link these terms ontologically, such that “knee replacement” and “knee arthroplasty” are mapped to unified concepts.

These approaches may also be extended into the social media environment, where we can measure the sentiments associated with pain-related terms expressed through a variety of online platforms. Sentiment analysis allows us to examine the context of such details within the sentiment of the underlying description.

Together, these techniques permit us to use raw text in machine learning experiments without resorting to manual extraction of data across thousands of documents.


Perioperative Social Network Analysis

Do given sets of healthcare professionals, by virtue of their teamwork, lead to differing outcomes following surgery? How can we quantify the complexity of communication systems inherent to modern healthcare interactions?

Our efforts in this realm employ social network analysis to examine the complexity, efficiency and redundancy of perioperative services at the network-level, especially those involving Acute Pain Medicine. (http://www.ncbi.nlm.nih.gov/pubmed/22568636?dopt=Abstract)


Systems-Level Acute Pain Medicine Optimization

In the last decade, anesthesiologists have significantly improved the efficiency and success-rate of regional anesthetics.

But how can we translate such technical improvements into hospital-wide improvements in acute pain management?

SLAP-MO explores how we can best incorporate the rapid, diverse improvements across a variety of acute pain medicine dimensions into safe, coherent, systematic improvements in the practice of acute pain medicine.(http://www.ncbi.nlm.nih.gov/PubMed/21697674?dopt=Abstract)