How does Arkstone validate its own data?

To ensure that the treatment recommendations provided by Arkstone are accurate and up to date, the OneChoice Decision Engine uses supervised machine learning overseen by our highly credentialed Quality Assurance Team.

Arkstone employs internal validation methods to ensure the accuracy and reliability of this machine learning processes. All datasets undergo human review before incorporation into the greater data model, or before the system "learns" what to do with the data. This meticulous review process is a vital aspect of Arkstone's QA protocol. Most importantly, data is never allowed to stagnate, and is periodically reevaluated within the QA framework to maintain consistency and rigor. This cyclical approach ensures the ongoing refinement and integrity of the machine learning process, even with familiar datasets.

Each data point sent to Arkstone undergoes immediate analysis by our computer system to confirm its relevance to the desired output and to ensure its recognition by the system. If the data is unfamiliar or bears resemblance but isn't an exact match to previous instances, it is flagged for additional review by our QA team to ensure meticulous parsing and organization.

During the QA process, humans train the OneChoice Decision Engine by correlating data points received, to clinically applicable guidelines. This is called Human in the Loop Validation. For example, data received may include age of the patient, organisms detected, resistance detected and allergies as well as sample type. Arkstone QA members will then correlate that data with well-established clinical guidelines and teach the machine what recommendations are best, based on those guidelines. This ensures that recommendations do not deviate from those standard of care guidelines. All the references used are then made available using OneChoice Plus so that healthcare providers may see the sources of the information for themselves.

While the ML process is designed to ensure computer accuracy, the QA process itself includes checks and balances to prevent human error as well. The Arkstone system includes safeguards to prevent the selection of incompatible treatment options. For instance, contraindications for pediatric or pregnant patients will prohibit QA members from inadvertently choosing medications unsuitable for these demographics. Furthermore, QA team members are unable to opt for treatment regimens not already approved for a specific diagnosis. This entails that the system strictly adheres to established FDA and treatment guidelines without deviation.

The validation process Arkstone uses incorporates variations and concepts from many well-known established modeling systems such as K-Fold Cross-Validation, Random Subsampling validation, Leave One Out Cross Validation and Hold out Validations as well as Human in the Loop Validations.

All this means that Arkstone's datasets are kept relevant and up to date, ensuring the best possible recommendations for providers and patients.