Theator FDA Letter to File_Sept 2021.docx

MEMORANDUM TO FILE TO: Theator File Date: TBD RE: Assessment to support Theator’s Surgical Intelligence Platform (also referred to as “Theator software product” or “Theator product”) is Non-Device Clinical Decision Support (CDS) software and is not subject to the applicable device requirements of the Federal Food, Drug, and Cosmetic (FD&C) Act. APPROVALS: Tamir Wolf Date Title: CEO Dotan Asselman Date Title: CTO I. Background A. Intended Use and Indications for Use Theator offers a real-world dataset software product (“the Theator product” or “the product”) comprised of three general modules for surgeons: (1) content management software; (2) informative intraoperative notifications; and (3) a visual operative report with analytics that couples what happens intraoperatively with patient outcomes to inform clinical management. The product captures, houses, and analyzes video data from surgical operations and is intended to be utilized by surgeons to review their surgical procedures in an efficient manner and generates insights that inform surgeons on their operative performance. The Theator product does not direct or drive clinical management and is not intended for use in making diagnosis or treatment decisions for patients. Surgeons can use the product to upload intraoperative footage of past surgical procedures, as well as accompanying data from patient records, collectively “content,” either manually or automatically. Content is then organized and displayed as a “library,” which allows a surgeon to search for videos of interest and display them. Theator’s software then combines computer algorithm capabilities, as well as human oversight and identification, to detect and identify steps of surgery, events of interest, key milestones and decision-making junctions (collectively, “flagged events”) in each video. All “flagged” events are specified in a video timeline within the Theator product for surgeons to view. The Theator product allows for additional customization where the surgeon can specify additional events of interest and pinpoint specific moments within a past procedure.

Preoperatively, surgeons can prepare ahead of surgery by issuing a query to view or review a procedure or certain types of procedures of interest. For example, in gallbladder surgery, events detected could include injury to a bile duct or to a blood vessel, or achieving a “critical view of safety,” or CVS. The surgeon can then view an identified decision junction; for example, a case in which a decision is made to insert (or not to insert) a drain towards the end of the procedure. In addition to search capabilities in Theator’s product to view past cases, curated content is generated for the operative surgeon prior to a case highlighting relevant previous cases and offering additional content which provides information to the surgeon of past performances. Postoperatively, a visual operative and analytics report (a “report”) in which what occured intraoperatively are identified and tallied to generate analytical statistics. These analytical statistics may include analyses of time of an operation or the duration of steps within, the number of events that occurred, the perceived complexity of a procedure, and comparison to other procedures, among other things. Surgeons receive the report following content upload, at which time they can review and analyze the video to learn from their performance and the surgical experience. As described in greater detail in Section IV below, the company believes that the product is Non-Device CDS software that informs clinical management and does not require premarket clearance or approval by the Food and Drug Administration (“FDA” or “the Agency”). Theator’s Surgical Intelligence Platform is intended for use by healthcare professionals (HCPs) who can independently review the basis of the recommendations and do not rely primarily on any recommendation provided by the software. Hence, the Theator product does not meet the definition of a medical device in accordance with the 21st Century Cures Act. Theator is memorializing its rationale for why the software does not fall within the scope of applicable device requirements of the FD&C Act in this memorandum to file. B. Product Description – Component and Functionality Description i. General Product Theator’s software is used by surgeons throughout their surgical workflow - preoperatively, intraoperatively, and postoperatively. Table 1: Theator’s Surgical Intelligence Platform software use cases Surgical Component General Function Workflow Step Pre-, intra-, and Content management - Allow upload, de-identification, storage, and post-operative and intelligent search organization of content - Automated recording, de-identification, and uploading of videos to Theator’s cloud platform - Allow search and querying 2

Preoperative Notifications - Tailored Preoperative preparation highlighting Intraoperative Livestream notifications relevant content for upcoming procedures Postoperative Visual operative report - Fully indexed, searchable library to find videos of and analytics interest - Viewable on any device, at any time - Set and allow customized, informative notifications to be sent to operating surgeon prior to procedure start - Allow one-way livestream of an ongoing/live procedure to another location - Allow review of past intraoperative surgical events during livestream for a non-operating healthcare provider - Set and allow customized notifications to be sent to non-operating healthcare provider at predetermined moments - See Table 2 below for a list of intraoperative notifications - Provide a highlight reel of surgical procedures such that even hours-long procedures can be reviewed within minutes showcasing key moments - Highlight reel to include annotations or flagging of procedure steps, intraoperative events, and safety milestones - Customize annotated timeline by pinning specific moments of interest and adding written or audio commentary - Collaborate with peers by tagging moments of interest with two-way communication via the product - Quantitative metrics to aid in the understanding of a procedure (e.g., procedure/step times, number of events such as leaks from vessels etc.) - Comparative metrics to individual surgeon’s performance and/or a group of surgeons, offering insight into surgical technique variability - Structured debrief/feedback tasks facilitated via Theator’s product 3

Postoperative surgical report automatically formulated based on intraoperative activity - Email digests highlighting key procedure moments and analytics - Validated assessment tools available for self- and/or peer-review available in the product Table 2: Theator’s Notifications Category Notification Livestream When reviewing live videos, all prior steps, events, and milestones can be (intraoperatively) reviewed, informing non-participating healthcare providers of the progress made in the surgery up to that point. Livestream Customized push notifications can be sent notifying the non-operating surgeon of (intraoperatively) outlier/aberant steps / events / milestones during surgery for viewing on the Surgical Intelligence platform or communicate the following notifications to the non-operative surgeon: ● The critical view of safety in a cholecystectomy has yet to be reached and the surgeon has been operating for [time determined by the hospital, e.g., 45min] ● The critical view of safety in a gynecology procedure has yet to be reached and the surgeon has been operating for [time determined by the hospital, e.g., 45min] ● The critical view of safety in a urology procedure has yet to be reached and the surgeon has been operating for [time determined by the hospital, e.g., 45min] ● Significant event detected ● Significant event detected and has not been controlled within a time span of [specified by hospital, e.g., 3 minutes] ● Procedure is taking longer (or shorter) than expected ● Procedure is estimated to begin in [x] minutes ● Procedure is estimated to be completed in [x] minutes ● Procedure converted to open from laparoscopy ● High complexity procedure has been identified ● Use of intraoperative cholangiogram detected in laparoscopic cholecystectomy All notifications are pushed to a non-operative surgeon to inform that surgeon on the status of the surgical procedure. OR efficiency OR scheduling team automatically notified of up-to-date procedure timing to (intraoperatively) inform them of an estimated completion time. OR efficiency Automatically notifies the operating surgeon at a specific time interval (e.g., (preoperatively) 10min) prior to the procedure start time to inform the surgeon of the estimated time when the patient is ready. ii. Content Management and Intelligent Search 4

Content is added subsequent to performing a procedure to an ever-growing repository of annotated, curated procedures (see Figure 1). The procedure videos are then matched via the algorithm in the software to that patient’s medical record, with identifiable patient information (such as MRN) only visible to the hospital employees based on predetermined permission levels. This enables search via specific video and/or patient characteristics, creating a highly searchable database. The library includes an intelligent querying and recommendation system. Querying can be performed for particular patient characteristics, videos from specific surgeons, videos for particular procedures, and videos at an intra-procedural level, including searching for occurrences such as procedure steps, intraoperative events, and safety milestones. These elements can be searched for on an individual level or combination and highlight the most relevant examples. Figure 1. Video content library as seen in Theator’s product showing example video collections (“My latest”, “Latest shared with me”, and “Featured”) One use case for such a library, or gallery, is reviewing past experiences ahead of a surgical procedure. The product offers a personalized experience by allowing for the showcasing of a variety of relevant videos and events of interest from similar patients. Today, surgeons leverage YouTube or other websites such as MEDtube, WebSurg, and Giblib in an effort to find relevant videos. Surgeons may either enter various parameters (e.g., patient demographics, prior medical and surgical history) by which to filter procedures, or they may enter a patient’s medical record number, and would 5

receive the most relevant videos to review. Theator’s product is intended to enable the surgeon access to his/her own procedure history, that of others in his/her department, or those from other centers, with similar or different experience levels. The library is composed of videos that have been automatically recorded in the operating room, de-identified, and uploaded to Theator’s cloud platform (videos can be manually uploaded, as well). Recording the surgical procedure occurs without any intervention required from the surgeon or surgical team, both to start and stop recording. Theator’s technology is visualization hardware agnostic, and can be applied to both laparoscopic and robotic visualization systems. The library is available to view on any device - web or mobile - at any time. Theator’s recommendation system, via its algorithm, also finds and displays “best practice” or benchmark procedures built from sequences of several procedures and creates a “rapid refresh” tailored to patient and provider, showing specific events that are more likely to occur in his/her upcoming procedure in order to improve preparation. This product is thus intended to be utilized so that surgeons can determine best practices in their review of previous surgeries. iii. Livestream notifications The basic capability of livestream is to enable one-way remote viewing of live videos at another location, accessed via Theator’s platform within the content library. In addition to livestream capabilities, Theator’s platform includes informative notifications. Informative notifications are available due to Theator’s computer vision software algorithm, which provides information to surgeons and surgical teams. Within the livestream capabilities, remote surgeons are able to review intraoperative procedural steps, events, and milestones that have occured during that procedure up to the current time. Several additional informative notifications are available, such as: customized push notifications informing a non-operating surgeon whether or not specific events or milestones have been achieved and estimated procedure start and end timing. Refer back to Table 2 above for a complete list of intraoperative notifications that are available only to non-participating healthcare professionals. iv. Postoperative reports and analytics Using the software algorithm, the product produces visual and written operative reports and analytics to allow for understanding trends and highlighting key takeaways. For example, content is analyzed, both by algorithm and by at least two human annotators, to generate procedure timelines. These timelines allow a surgeon to easily “dive” into specific sequences of interest within a surgical procedure without the need to review the entire operation. This functional ability has significant value as it enables debrief and feedback for the operative surgeon: an actionable methodology to rapidly review each procedure. The visual summary is an automated creation of a highlight reel based on predetermined characteristics, or discrete user preferences. For example, a 4-hour long laparoscopic partial kidney resection may be reduced to a several-minute-long video (“highlight reel”) composed only from predetermined relevant steps, and events.  The entire video is also available for the surgeon to view. 6

Both the full video and the highlight reel can be watched 0.5-8x speed. Editable versions can be made within a surgeon’s content library to further customize the video into segments of interest. Figure 2. Video summarization and timeline, listing procedure steps, events, milestones, and key moments of interest By assessing the procedural footage, the computer vision software can derive meaningful moments that occurred within a procedure to populate a written post-operative report. Additionally, the software can also derive surgeon-specific analytics, which include key parameters such as overall procedure time and/or intra-procedural step durations, number of ab/normal events that occurred, assessment of surgeon technical skill, complexity grade of the procedure, surgical technique variability, and so forth. These analytics can also be viewed across a group of surgeons. Leveraging analytics over time may aid surgeons and others, such as surgeon leadership, in understanding capabilities and utilizing such documentation to further enhance surgical performance. 7

8

Figure 3. Illustration of general surgery analytics. Top to bottom: procedure breakdown across general surgery for the individual and department, procedure complexity and event breakdown for the individual and department, procedure duration for appendectomy comparing individual vs. department. Procedure information captured, such as key moments and analytics, are curated into email digests that are sent to users postoperatively. These are personalized for the stakeholder and are customized to highlight meaningful insights. In addition to procedure analytics, validated assessment tools, such as Global Evaluative Assessment of Robotic Skills (GEARS) and Global Operative Assessment of Laparoscopic Skills (GOALS) are available within the product for self- and/or peer-review grading of overall technical proficiencies. These tools are not developed by Theator, rather offered within the platform that can be used to measure and improve proficiency of surgeons through education and training. Figure 4. Validated assessment tools example 9

iv. Annotations In order to create visual operative reports and provide accompanying analytics, various aspects of procedure videos are annotated using two different methods: a semi-autonomous annotation system and a human oversight and annotation system. Our annotation system is composed of several components, as depicted in the workflow diagram in Figure 5, below. Figure 5: Workflow Diagram Legend ● Grabber [software/device] – “grabs” recorded video from the operating room, acquires server info and sends to our system flow, the gateway to the processing queue. The grabber could also be our web-app in a manual upload process; in the future we will have a module that handles automatic integration with operating room systems. ● Processor - receives its tasks from the grabbers, runs our main algorithms, sets annotations and metadata autonomously. ● Human annotator - receives semi-annotated procedures from the processor, fine-tunes and adds his/her own annotations. In case of unknown events the annotator creates a task and pushes it to the “unique cases queue.” ● Human validator - receives fully annotated procedures from the human annotator, validates the annotations, fixes mistakes and adds missing annotations. 10

● Data warehouse - store all the videos and annotations. ● Model trainer - module that receives new video annotations and trains/fine-tunes the Processor’s algorithmic models to continuously improve their accuracy using new annotated data. ● “Expert” annotator - human annotators with various levels of surgical training oversees annotation of unknown/unique events, adds new event types and annotates more complex events. ● “N ✖” or “✖ N” - more than one Our semi-autonomous annotation system extracts key information from raw, unstructured surgical footage. There are several types of annotations currently performed: procedure steps, intraoperative events (both desired and undesired), safety milestones, and procedure complexity – see Figure 6. In some cases, the software also annotates key attributes of an annotated object. For example, when annotating an organ, the software annotation process might include annotation of the level of inflammation of the organ. As another example, in gallbladder surgery (e.g., laparoscopic cholecystectomy) there is a certain event called “critical view of safety,” which is considered a requisite for performance of a successful procedure. The quality is annotated by using our software. Figure 6. Examples of annotations For illustration purposes, analytics based on such annotations include, among others, procedure complexity, tool utilization, timing, errors, and complications, all considered in Figure 7 as metrics. 11

Figure 7. Annotations and metrics These annotations and metrics built upon raw surgical video data can then be combined with patient and surgeon information to generate indexed procedures that are then housed in the surgical video library. See Figure 8 for more details. 12

Figure 8. Building indexed procedures from raw data In some cases, a step may include a unique event, one which has not been encountered previously, that is not annotated/indexed and is neither recognized by our algorithmic models nor human annotators. Since steps in procedures are detected automatically, these “abnormal” ones can be marked by the algorithm for the human annotator to conduct a more thorough investigation of a unique event occurring within the phase. The annotation process is accomplished in a hierarchical (hybrid) manner, starting with autonomous annotations, a manual annotator, and in some cases, surgeons that address more complex scenarios (see Figure 9): 13

Figure 9. Annotation workflow for difficult cases Each procedural video is first annotated automatically and available for review immediately postoperatively. Then, each video undergoes rigorous analyses by at least two different human annotators to confirm the automated annotations were assessed correctly, updating the annotations as necessary. The first performs validation of our algorithmic autonomous annotations, while the second adds an additional layer of validation. As additional data is gathered, the automated annotation algorithms mature and accuracy levels continuously increase. Theator’s software is intended solely for informational purposes and is not intended for medical monitoring purposes, for diagnosis, for determination of the state of health of a patient, nor for cure, mitigation, treatment, or disease prevention. II. Legal / Regulatory Background 14

Section 201(h) of the Federal Food, Drug, and Cosmetic Act (“FDC Act”)1 defines a medical device as: an instrument, apparatus, implement, machine, contrivance, implant, in vitro reagent, or other similar or related article, including any component, part, or accessory, which is— (1) recognized in the official National Formulary, or the United States Pharmacopeia, or any supplement to them, (2) intended for use in the diagnosis of disease or other conditions, or in the cure, mitigation, treatment, or prevention of disease, in man or other animals, or (3) intended to affect the structure or any function of the body of man or other animals, and which does not achieve its primary intended purposes through chemical action within or on the body of man or other animals and which is not dependent upon being metabolized for the achievement of its primary intended purposes. Section 3060(a) of the 21st Century Cures Act amended the FD&C Act to add section 520(o) of the FD&C Act, which excludes certain software functions from the definition of device in section 201(h) of the FD&C Act. Certain CDS software functions are excluded from the definition of device by section 520(o)(1)(E) of the FD&C Act. Specifically, this section excludes, from the definition of device, software functions that meet all of the following four criteria: (1) not intended to acquire, process, or analyze a medical image or a signal from an in vitro diagnostic device or a pattern or signal from a signal acquisition system (section 520(o)(1)(E) of the FD&C Act); (2) intended for the purpose of displaying, analyzing, or printing medical information about a patient or other medical information (such as peer-reviewed clinical studies and clinical practice guidelines) (section 520(o)(1)(E)(i) of the FD&C Act); (3) intended for the purpose of supporting or providing recommendations to a health care professional about prevention, diagnosis, or treatment of a disease or condition (section520(o)(1)(E)(ii) of the FD&C Act); and (4) intended for the purpose of enabling such health care professional to independently review the basis for such recommendations that such software presents so that it is not the intent that such health care professional rely primarily on any of such recommendations to make a clinical diagnosis or treatment decision regarding an individual patient (section 520(o)(1)(E)(iii) of the FD&C Act). 1 21 U.S.C. § 321(h). 15

III. Application of FDA’s Policies to Theator’s Surgical Intelligence Platform There are three primary functions of the product being contemplated: first, the ability to store and view archived videos from surgical operations; second, intraoperative informative notifications to the surgeon and surgical team (see Table 2 in Section I above for the list of intraoperative notifications); and third, the algorithm is intended to identify flagged events that occurred during surgery to generate surgical analytics. All of these software functions are non-device CDS and meet the criteria to be excluded from the device definition under section 520(o) of the FD&C Act. The Theator Surgical Intelligence Platform does not acquire, process or analyze any medical images or patterns or signals from a signal acquisition system. The videos from surgical operations, intraoperative notifications and surgical analytics are intended to inform HCPs with additional insights of what is happening intraoperatively. The instructions for use (IFU) describe in plain language the intended use of the software function, the intended users, the inputs and logic used by the algorithm and the basis for rendering a recommendation. HCPs can independently review the basis of the recommendations and do not rely primarily on the software’s recommendations. They use the information provided by the Theator Surgical Intelligence Platform as one of several data sources to develop their own, independent decision on course of action for patients. As further explained in the draft FDA guidance document, “Clinical Decision Support Software”, dated September 27, 2019, software used to inform HCPs of a healthcare situation where the user can independently review the basis of the recommendation do not meet the definition of a medical device and are Non-Device CDS functions. IV. Conclusion At this time, the Theator Surgical Intelligence Platform does not meet the definition of a medical device per Section 201(h) of the FD&C Act and Section 3060(a) of the 21st Century Cures Act. The regulatory classification of the product may change as additional features are added in future. Theator will continue to evaluate these additional features and revise this memorandum to file to document any updates to the regulatory classification, as required. 16