By Chris Plance and Kunal Patrawala, healthcare experts, PA Consulting.
Artificial intelligence tools, such as natural language processing (NLP), can be applied to unstructured data to produce real-time and nuanced insights. NLP can be applied to many types of unstructured data in a provider setting, however, the ideal approach is to apply these tools to a non-clinical source of unstructured data first.
Non-clinical unstructured data provides organizations with a perfect platform to develop their skills. Two well-known sources of such non-clinical unstructured data are consumer assessment of healthcare providers and systems (CAHPS) and hospital consumer assessment of healthcare providers and systems (HCAHPS) surveys. Applying NLP to CAHPS and HCAHPS surveys can help providers improve their top-box scores, build confidence and capabilities in their data skills, increase revenue, and allow organizations to take the crucial first steps towards the journey of unlocking 100% of their data.
CAHPS and HCAHPS Surveys play an important role in today’s Healthcare environment and are a substantial source of structured and unstructured data
CAHPS and HCAHPS are rating scale-based surveys that help providers discern patient perspectives such as patient experience and satisfaction. In 2019, more than 3 million patients completed HCAHPS surveys. Patient experience metrics from HCAHPS surveys create 25% of Hospital Value-Based Purchasing total performance scores, while CAHPS surveys are used by providers enrolled in the Merit-based Incentive Payment System (MIPS) program.
Similarly, private payers are also tying these metrics to reimbursement. The advent of healthcare consumerism has also led to patients extensively using tools such as CMS’s Star Ratings to make informed decisions. As a result, CAHPS and HCAHPS surveys have downstream effects on reimbursement, star ratings, and brand loyalty which makes improving top-box CAHPS and HCAHPS scores vital to a provider’s financial and operational health.
CAHPS and HCAHPS contain both structured (numerical ratings) and unstructured data (patient comments). These patient comments are accessible, and approximately 50% of the patients who take HCAHPS surveys leave comments. Unlocking this unstructured data can provide more comprehensive and specific information that cannot be gauged from only structured numerical ratings.
The current approach of analyzing unstructured data is qualitative and labor-intensive
Despite the importance and value of these comments, they are underutilized because providers usually hire quality improvement abstractors to manually evaluate every patient comment. This makes analyzing unstructured data a labor-intensive, time-consuming, qualitative, and ultimately expensive process. However, with the recent advances in machine learning and artificial intelligence, analyzing unstructured data can be automated using NLP.
NLP can produce real-time insights such as drivers of patient satisfaction and rationales of poor patient experience from unstructured patient comments, which would help improve top-box HCAHPS and CAHPS scores at a significantly lower cost compared to traditional methods.
NLP can unlock unstructured data in CAHPS and HCAHPS surveys through automation
Everybody has had some exposure to NLP thanks to its ubiquity and wide-ranging applications, from spam filters and spell checkers to language translators, chatbots, and voice assistants. NLP is a method which allows computers to interpret and analyze unstructured human language by combining artificial intelligence and linguistics. NLP interprets unstructured language by applying algorithms to identify patterns which offer valuable insights.
NLP can break down patient comments through various techniques, such as topic classification and sentiment analysis. Topic classification can help organize unstructured patient comments into topics such as patient-physician communication, cleanliness, and wait times. This allows providers to focus on comments that relate to specific patient experiences, by filtering for relevant topics.
Sentiment analysis can further filter the data based on emotional sentiments such as positive, negative, or neutral. The combination of different NLP techniques exponentially improves the analysis by making it more nuanced. These nuances can help providers identify the specific rationales and patterns driving middle and bottom-box CAHPS and HCAHPS scores.
These patterns can provide actionable insights allowing healthcare professionals to first diagnose underlying issues, perform root cause analysis based on data driven insights, and then create targeted strategies to improve middle and bottom-box scores.
The implementation of these tools, though, is not as simple as purchasing a piece of software from a vendor. Organizations must develop skillsets to better prepare data and understand the possible outcomes, alter workflows and decision-making processes to become data-driven, and ultimately empower staff to act on these insights. An example of this complexity is illustrated by the fact that 30.5% HCAHPS surveys with positive ratings have unstructured negative comment. However, a sophisticated partner can guide organizations through this journey and help them unlock 100% of their data.
There are several success stories across the healthcare industry
Providers have already started using NLP to unlock their unstructured data. New England Baptist Hospital, University of Utah Health, and Harris Health System have all used NLP algorithms in some capacity. New England Baptist Health is using NLP to analyze negative patient comments, from HCAHPS surveys, after a patient had undergone a total shoulder arthroplasty.
They were able to recognize that many negative patient comments were related to factors such as the temperature of the room, lack of seating for family members, and discharge delays. This information would be impossible to discern from structured CAHPS and HCAHPS data. The University of Utah Health applied NLP to more than 50,000 HCAHPS surveys with similar results.
Their analysis indicated appointment wait times and appointment access had the highest number of negative comments. Harris Health System (HHS) has been using NLP to analyze feedback received from its real-time feedback program. NLP was able to identify that 44% of the negative comments were directed towards wait times. This NLP algorithm was also able to discover patients were more dissatisfied with the lack of communication around the delays than the delays themselves.
Organizations should invest in NLP to improve top-box scores and begin building their data muscles
NLP is a powerful tool that can equip providers with superior information and provide actionable insights not readily available through structured data alone. These insights can help providers make highly impactful data-driven decisions resulting in improved patient experiences and top-box CAHPS and HCAHPS scores at a cost significantly lower than orthodox methods.
Leveraging NLP to unlock the unstructured data found within CAHPS and HCAPHS provides a highly impactful solution, in a short time frame, without the need of using any clinical data. In addition, it also creates a perfect launchpad to begin accessing the 80% of data which is currently unutilized. NLP can create quick wins when applied to non-clinical data, however, significant opportunity to improve patient outcomes, increase revenue, and reduce costs lies in applying an organization’s new NLP skills to clinical data.