ORIGINAL ARTICLE Neural network-supported patient-adaptive fall prevention system Mehmet Hilal O ¨ zcanhan 1 • Semih Utku 1 • Mehmet Suleyman Unluturk 2 Received: 28 February 2019 / Accepted: 19 August 2019 Ó Springer-Verlag London Ltd., part of Springer Nature 2019 Abstract Patient falls due to unattended bed-exits are costly to patients, healthcare personnel and hospitals. Numerous researches based on up to three predetermined factors have been conducted for preventing falls. The present comprehensive proposal is based on four sub-systems that synthesize six factors. A parameter is assigned to each factor with a coefficient specifically determined for each individual patient and per admittance. The parameters are aggregated in equations that lead to an early warning about a probable bed-exit, or an alarm about an imminent bed-exit. The ultimate aim of our proposal is the generation of the earliest possible warning to grant the longest time for nurse intervention. Thus, the probable fall of high-risk patients can be prevented, by stopping the unattended bed-exits. The proposal is supported by a prototype multi-tier system design and the results of laboratory patient bed-exit scenarios, carried out using the design. Comparison of the obtained results with previous work shows that our proposed solution is unmatched in providing the longest time for nurse intervention (up to 15.7 ± 1.1 s), because of the comprehensive six-factor synthesis, specific to each individual patient and each admittance. Keywords Fall prevention  Medical systems  Patient safety  Wearable sensors 1 Introduction The Joint Commission International (JCI) has announced that patient falls are described as ‘‘unplanned descend to the floor, resulting in an injury or not.’’ According to JCI, patient falls constitute 10–14% of sentinel events [1]. Falls ending with injuries limit future patient activities require prolonged medical care and sometimes cause deaths [2]. In essence, falls are reported to cause depletion of patient, hospital and society resources [3]. For example, hospital stays are extended by 14–34 days [4]. In numbers, the total cost of falls in 325 hospitals has been estimated at 16 million dollars [5]. A closer study of falls shows that the majority of falls are unwitnessed, as the patients are reluctant or unable to call for assistance [6]. Some unwitnessed falls that go undetected for a long time are detrimental to the patients and healthcare personnel [6, 7]. A study shows that falls occur especially in the surroundings of the bed. All bed- side falls are classified as bed-exit falls [8]. Overall, future fall numbers are predicted to increase in parallel with the increasing number of aging people [9]. As a counter measure, governments have made prevention of hospital falls a priority of their national healthcare policies and the researchers have stepped up their efforts in devising methods for prevention, detection and mitigation of falls [10]. Although there are many different fall mitigation methods, none of the proposed directly prevent falls. The methods can be categorized under four groups: 1. Using restraint devices (e.g., side rails, patient restraint belts, etc.), 2. Placing electronic sensing devices (sensors) in and around the patient room (e.g., atmospheric room & Semih Utku semih@cs.deu.edu.tr Mehmet Hilal O ¨ zcanhan hozcanhan@cs.deu.edu.tr Mehmet Suleyman Unluturk mehmet.unluturk@yasar.edu.tr 1 Department of Computer Engineering, Dokuz Eylul University, Izmir, Turkey 2 Department of Software Engineering, Yasar University, Izmir, Turkey 123 Neural Computing and Applications https://doi.org/10.1007/s00521-019-04451-y