KOAD-ICU

KOAD-ICU is a research tool that addresses the lack of effective patient monitoring framework in local hospital ICUs in Bangladesh. In a typical ICU setup, nurses conduct round the clock monitoring and manually record patient data on tabular sheets. Specialist physicians are notified in the events of emergencies. For understaffed ICUs, front-line caregivers must work tediously long shifts and it is common for critical events to go unnoticed. This increases the risk of injuries and death of patients. 

To that effect, we designed KOAD-ICU, a centralized, automated patient monitoring and real-time emergency signaling system. We developed a working prototype that comprises of inexpensive peripherals and simplistic user interface. The modular design is amenable for use within any medical setup. KOAD-ICU keeps continuous patient records from monitoring units, analyzes patient data online, and includes intelligent alarming features. The system can be utilized to research sustainable patient surveillance approaches in hospitals.

In the last four decades, total number of ICU beds in hospitals has increased  4000% (approximately, from 28 in 1980 to 1200+ in 2019). However, ICU workforce data published in multiple studies on Bangladeshi hospitals indicate significant shortage of staff. 

Commercially available central monitoring systems offer optimization of human, financial, and technological resources. However, these systems are complex, require training, and cost of maintenance is very high. Local hospitals have opted for reactive model of care to cut costs. Such patient care is a combination of scheduled visits from intensive care nurses and alert-responded consultations by specialist physicians. This approach has contributed to an innate alarm fatigue among the nurses. They keep manual records of patient vital statistics from various bedside devices on tabular sheets. Manual records are unreliable: they can introduce inaccuracy from incomplete entries; they are often difficult to read; and they may be lost or otherwise become inaccessible. 

Yet international practice guidelines, clinical protocols, and operational references emphasize the necessity of digital patient database and integration of medical equipment. 

The purpose of intensive care is to ensure patient's survival by preventing further physiologic deterioration. Our developed system can help physicians and nurses to treat and resolve the underlying disease of ICU patients efficiently. 

KOAD-ICU benefits patients with immediate and responsive services from ICU staff:

• Hospitals will have access to summary reports and digital medical records for each patients. The nurses who had to tabulate readings can invest in other aspects of patient care.
• Automation will decrease false alarm rates. Front-line caregivers need not endure alarm fatigue. 
• The centralized surveillance architecture will eliminate any hidden irregularities in patient monitoring. Physician oversight is augmented instead of being replaced.

Our working prototype integrates a data extraction adapter, a transceiver computer, a router, a KOAD computer, and a central monitor. The KOAD patient surveillance application extracts data from PMU and inserts into KOAD database. Then the program processes patient data by adopting machine learning model, specifically Kernel-based Online Anomaly Detection (KOAD) algorithm. The algorithm clusters inherently correlated biological measurements together and alerts when there is a break in the pattern. The central monitor is continuously updated to display the underlying situation of each patient connected to our surveillance system.  

For our experiments, we utilized a supervised cross validation approach. Also, we scrambled any identifiers in the patient data to preserve anonymity. Patients were not exposed to any risk during the course of our study. 

We intend to modify the prototype further to accommodate some useful features, such as: 

• reconfigure the application so that ICU staff require minimal training to operate
• construct an interactive display that can be customized according to ICU capacity
• store accurate digital records securely

KOAD-ICU can process the massive quantities of patient surveillance data according to physicians' specifications. The objective of our solution is to provide data management ability for the ICU staff. 

Typical bedside PMUs continuously display the physiological data of patients. The existing PMUs fail to address this inherent correlation between the biological signals. They sound warnings when the built-in change point detection algorithm detects deviation in any of the parameters. In reality, the devices bleep continuously which contribute to innate alarm fatigue among the nurses. 
On the contrary, KOAD algorithm: 

1. evaluates the complicated, underlying structure and signals when there is a break in the pattern 
2. detects anomalous events in real time with high accuracy and low false alarm rates 
3. requires lightweight computational and memory resources

ICU teams in local hospitals will be aided with ordered patient monitoring and emergency signaling system.

In order to construct KOAD-ICU, we took a step by step approach to familiarize with intensive care facilities. We consulted studies that highlighted key surveillance issues. We surveyed a few ICUs to understand the setup and model of care in each one. We procured ethical clearance from one hospital to extract physiological data from their PMUs on the basis of protecting anonymity of the subjects. 

To evaluate the performance of KOAD-ICU, we monitored 5 patients continuously for 8 hours. For instance, in a dataset of total 2000 timesteps, after running KOAD with emergency events manually identified by a ICU specialist on first 300 timesteps, the system was able to correctly identify 27 out of 36 anomalies in subsequent 1700 timesteps. 

The scope KOAD-ICU does not include inferring whether patient condition has improved. We are focused to provide hospitals with accurate digital records, decrease alarm fatigue in ICU staff, and ultimately install our system as the primary central monitoring framework in local hospitals. 

Our prototype is installed in 5 beds of a 16-bed ICU in a hospital in Dhaka. We are aiming to complete our working prototype to provide comprehensive automated signaling for the hospital. KOAD-ICU assists 20 ICU staff currently. Within next year, we expect to serve minimum 500 ICU staff and in five years, the solution will aim to serve 10,000 ICU staff at least.

Within the next year, our scope of implementation are:

1. full application development 
2. run field-level measurements on cloud 
3. obtain proprietary patent(s) 
4. prepare training guidelines, obsolescence statement 

We are targeting to achieve the following goals within the next five years: 

1. successful completion of alpha and beta tests 
2. receive investment funding from short-listed donors 
3. production and distribution of KOAD-ICU 

Our research project requires considerable funding for continuity of work, salaries, materials and supplies, showcasing and more. Additionally, for impressionable effect in future, we are short of detailed design, work plan, showcasing and installation. There is also insufficient awareness in our healthcare system for implementation of solutions such as ours. We are yet to receive ample assistance from healthcare providers regarding access to ICUs. 

Currently, we rely on our institution to provide yearly grants for research assistant salaries and materials. Since the amount is overall unsustainable, our team pursue external grants for these financial barriers. We plan to appoint full-time Research Lead and Project Manager, hire legal experts to file patent and other documentation, and employ related professionals to draw detailed design and prepare work plan. The technical barriers such as absence of system-level design and laboratory setting can be overcome by consulting guidelines and suppliers. 

Sponsored research grant approved research group in Department of Computer Science and Engineering, Independent University, Bangladesh.

Principal investigator: one 

Co-investigator: one 

Research assistant: two

Total: four

Dr Tarem Ahmed, the principal investigator, has developed KOAD algorithm. He has supervised research groups implementing KOAD in automated visual surveillance, network flows in high-speed backbones and hospital ICU implementation. 

Dr Faisal M Uddin has expertise in areas such as communication network design, operations research/management. He is the co-investigator. 

Ms Nazifa Mubashshera Shemonti has completed her undergraduate thesis on KOAD in hospital ICU and has developed the concept for KOAD prototype. 

Mr Shifat Uddin has been working in the KOAD team for almost two years. He consults on network and hardware implementation. 

We have ethical clearance from a hospital in Dhaka where we have set up our prototype.

Customer segments: local hospitals 
Value proposition: secure patient database; reliable and customized report 
Channels: Showcasing 
Customer relationship: on-site technician 
Revenue stream: subscription for customized reports 
Key activities: development and manufacture; distribution and logistics 
Key resources: relevant personnel on payroll or consultation; technology 
Key partnerships: head of departments in hospitals; donors 

Short term: sustained donation and grants 

Mid term: raise investment capital and donations 

Long term: sell services 

The viewpoints of KOAD-ICU perfectly orient with the focuses of the Tiger IT Foundation. Our product integrates proven data mining, machine learning, and pattern recognition techniques. Also, we employed engineering skills to develop KOAD-ICU to improve ICU patient monitoring. We are confident that, by harnessing the mentorship opportunity, our product can reach intended outcomes if KOAD-ICU receives the Tiger Challenge prize funding. 

• Healthcare providers 
• Medical equipment manufacturers