We are actively searching the healthcare industry for partners to help drive Philter’s development. Please contact us if you or your organization is interested.
Philter is an application that analyzes, reports on, and removes Protected Health Information (PHI) from natural language text. Philter is capable of removing PHI information such as phone numbers, email addresses, social security numbers (SSNs), patient names, and locations from text. Philter was created for environments in which PHI text is utilized within defined boundaries and was designed for big-data environments but supports flexible deployments into virtually any environment whether cloud or on-premise.
PHI data breaches can be very costly and in the millions of dollars in fines and even more in damage to reputation. Philter can play an important role in helping to keep PHI from crossing disallowed boundaries in your HIPAA-controlled environments.
Capabilities of Philter
- Identification and removal of PHI from text. Philter uses natural language processing to identify persons, places, and things for removal in text.
- Detection of PHI in text. (Answers the question, “Is there PHI in my text?”) This capability involves analyzing your text to assign it a numeric score that indicates the likelihood of PHI.
Philter can be launched immediately on Amazon Web Services and Microsoft Azure via their respective marketplaces. For other deployments or managed services please contact us and we can assist. Please note that prior to using Philter in a HIPAA-controlled environment additional configuration must be done in order to make Philter compliant. Refer to this guide or contact us for more information.
How Philter Works
Philter works by analyzing the text it receives for PHI. Some PHI is easily recognizable through patterns, such as social security numbers and zip codes. Pieces of the text matching the pattern are replaced by user-configured placeholder text. Detecting other types of PHI is more complex because it does not follow a set pattern. These are items such as patient names and locations. Philter identifies this information by using trained models created to detect these items. Philter can also report on PHI in text without removing it.
Philter is capable of learning and improving its accuracy over time. Filtered names and locations that have been identified in text can be applied to future text.
Please note that model-based detection is not an exact science and Philter’s accuracy and performance should be evaluated against your data prior to a production deployment. Adjusting Philter’s configured sensitivity can have a significant impact on the accuracy.
Philter can be used in multiple ways. Which way is best depends on your environment and your use-case. For help deciding please get in touch. First, Philter exposes an API that receives raw text and returns the filtered text. Second, Philter supports reading raw text from an Apache Kafka topic and writing the filtered text back to a separate Apache Kafka topic. Lastly, Philter can be used from the command line.
Designed for Efficiency
Anytime a new process is injected into an existing process there is an effect on performance. We recognize that and we designed Philter to have as little performance impact as possible. For additional processing capabilities, Philter can take advantage of GPU-capable systems by utilizing the GPU during its analysis.