MedSense detects opportunities for hand hygiene in four steps: (i) badges detect "events" when the HCW moves in and out of patient zone; (ii) events are assigned a probability of patient contact based on duration; (iii) events with high probability of patient contact are split into "Before Touching a Patient" and "After Touching a Patient" hand hygiene indications; and (iv) isolated indications are counted as opportunities while "After Touching a Patient" indications followed by "Before Touching a Patient" indications in quick succession are combined into single opportunities.
The system defines an "event" as an interval when a badge-wearing HCW spent time in a patient zone within range of a Beacon installed on the wall at the head of the patient's bed. The Beacons, which focus their transmissions into an elliptical field around the bed, periodically broadcast, and the Badges receive these "pings" and record the patient zone ID and signal strength. The "Received Signal Strength Indication" (RSSI) of the ping functions as an indicator of distance between the two devices. During the technical phase of the trial, the Beacons were calibrated such that a patient zone extending approximately arm's length from the bed's perimeter could be detected by applying a threshold to the RSSI (Figure 6). A detection algorithm inputs these ping data points and calibration values and outputs a series of events defined by start and stop times together with the patient zone and badge ID. The algorithm uses a timeout of one minute where a badge may leave the patient zone and return while continuing the active event.
Patient Contact Inference
MedSense uses a predefined reference table to predict the probability of patient contact having occurred during an event. The table is indexed by event duration, and the corresponding probability value represents the probability of patient contact during the event. The table's values derive from the results of data observation on the unit, which showed a strong relationship between the duration spent in a patient zone and patient contact occurring. Figure 7 shows the probability of patient contact in relation to the event duration. Events with a low probability of patient contact (duration less than fifteen seconds) are disregarded, and the remaining events each create two indications for hand hygiene: "Before Touching a Patient" and "After Touching a Patient", which are assigned times equal to the start and end times of the events, respectively. In addition to type and time, the indications carry forward their probability of patient contact as a weighting factor to be used in the compliance calculation.
According to the WHO's recommendation, the occurrence of a single indication creates an opportunity for hand hygiene. MedSense therefore counts each isolated indication as an opportunity with probability of occurrence equal to the indication probability. When multiple indications occur at the same time, the WHO specifies that only a single opportunity should be counted. MedSense groups an "After Touching a Patient" followed by a "Before Touching a Patient" indication that happen within two minutes of each other as a single opportunity. When combining these two indications, the resulting opportunity has a probability equal to the probability that at least one of the two indications occurred.
Wireless sensors detect when HCWs dispense alcohol and soap product, and then broadcast an activation message to proximate badges, which record the messages. The action algorithm selects activation messages with strong signal strength and assigns them to badges as hand hygiene actions.
Pump bottle sensors accommodate a single alcohol or soap bottle. The sensors can be mounted on a wall or placed on a flat surface. When the HCW presses on a bottle's pump to dispense product, a force sensor module in the bottom of the unit triggers and broadcasts a message indicating that a pump bottle "activation" occurred.
Each badge that receives a particular activation message records the identifying information together with the time and RSSI. When this data is uploaded to the server, the action algorithm selects activations with an RSSI above a threshold as representing badges, and therefore HCWs, who could have initiated the hand hygiene action. When a single activation is selected for a particular action, the action algorithm directly assigns it to the corresponding badge. If multiple activations are selected, the algorithm assigns an action to each represented badge but with a flag marking them as uncertain.
The compliance algorithm calculates compliance from a set of opportunities and actions in three steps. The first step involves matching the actions to opportunities based on temporal proximity. In the second step, the algorithm filters out the opportunities matched to uncertain actions. Finally, the matched and unmatched opportunities feed into a calculation that determines the compliance.
Matching Actions to Opportunities
The matching algorithm uses the following criteria to determine which actions match to which opportunities: (i) an action can only match to a single opportunity; (ii) multiple actions may match to the same opportunity; (iii) a matching action and opportunity must occur within 90 seconds of each other; and (iv) an action cannot match an opportunity if there is an intervening opportunity. The algorithm determines each match in order from shortest to longest time between action and opportunity. When an action marked as certain (from the action algorithm) matches to an opportunity, the algorithm removes the opportunity from the potential match set so that no additional actions may match it. The end result is three types of opportunities: (i) no action matched; (ii) certain action matched; (iii) one or more uncertain actions matched. Figure 8 illustrates the matching algorithm.
Filtering Out Uncertain Matches
The subset of the opportunities matched to uncertain actions represents a case where the system did not have the discriminatory power to determine compliance behavior. To reduce error in the final compliance calculation, the compliance algorithm filters out these ambiguous data points. The remaining opportunities, those with certain actions or no action at all, are referred to as compliance data points.
MedSense defines the compliance as the conditional probability of an action given an opportunity, denoted P(A|O). P(A|O) is equivalent to the joint probability of action and opportunity divided by the marginal probability of an opportunity, or P(AO)/P(O). Since the algorithm has removed the uncertain actions, this calculation becomes a weighted average where the action outcomes (zero or one) are weighted by the opportunity probabilities. This compliance calculation can be performed over any subset of the compliance data points, as is the case when calculating a compliance for a window of time, a category of HCWs, or any other grouping variable.