Verification using the eye-tracking system

Eye-tracking, a method of verifying gaze movements by detecting the corneal reflex of infrared rays, is used in various fields such as medicine, psychology, and cognitive science [23,24,25,26]. In this study, we investigated the gaze movements of pharmacists in the dispensing process using a glasses-like eye tracker (Tobii Pro Glasses 3, Tobii Technology K.K.). Gaze movements obtained by eye-tracking were mainly classified into the two categories fixation (stagnation within a 20-pixel window for a minimum of 100 ms) and saccade (quick movements of the eyeballs). Fixation and saccade were judged from recorded motion videos using dedicated analysis software (Tobii Pro Lab Analyzer, Tobii Technology K.K.).

Target persons and drugs

The inclusion criteria for pharmacists in this study were as follows. First, an essential criterion required for accurate eye movement measurements was that pharmacists should be able to read the dispensing information displayed on the large monitors with their naked eyes or while using soft contact lenses. Second, pharmacists should have more than 18 months of dispensing experience at the Kyushu University Hospital; this was essential to maintain the quality of verification above a certain level. Finally, pharmacists should agree to participate in this study.

The target drugs used in this study were 15 pairs of same-name drugs dispensed in the hospital. Here, “same-name drug” refers to a pair of drugs with the same name (character part) but a different ingredient quantity (number part). The target drugs were tablets of famotidine OD 10 mg/20 mg, nauzelin® OD 5 mg/10 mg, forxiga® 5 mg/10 mg, eliquis® 2.5 mg/5 mg, zolpidem 5 mg/10 mg, furosemide 20 mg/40 mg, cilostazol OD 50 mg/100 mg, atorvastatin 5 mg/10 mg, depakene® R 100 mg/200 mg, and decadron® 0.5 mg/4 mg, furthermore, the former drugs in five pairs of ones are as follows: losartan K 25 mg/50 mg, Topina® 50 mg/100 mg, tegretol® 100 mg/200 mg, belsomra® 20 mg/25 mg, and thyradin® S 25 µg/50 µg.

Preparation of the verification slides

The slides used for dispensing verifications in this study were created using Microsoft PowerPoint® 2016 and each dispensing verification was performed as a set of one prescription slide and three drug rack slides. Five target drugs were dispensed in each verification. The dosage and administration of each target drug were appropriate and there were no drug interactions among the five target drugs.

Regarding the content of a prescription slide, basic information consisting of patient name, age (sex), body weight, height, and creatinine clearance value was displayed at the upper side of the slide. Moreover, the dispensing information consisting of four items namely (a) drug name, (b) drug usage, (c) location display, and (d) total amount was displayed in the center of the slide. Regarding the content in each drug rack slide, a grid-type rack of 5 rows × 10–14 columns was displayed, each cell containing a name label of a drug at the bottom. A total of three drug rack slides were prepared in each verification, and the five target drugs were arranged at specified positions on them. Drugs with the same initial character and ingredient quantity (number part) were not displayed in the same row as the target drug. Additionally, the arrangement of drugs on the verification slides differed significantly from the actual dispensing state at Kyushu University Hospital. There was no adherence to either alphabetical order or order based on drug efficacy. Thus, approximately 180 drugs including the five target drugs were displayed on the three drug rack slides.

In this study, the indication method of “(c) location display” in the prescription slide was classified into two types: “numeral combination” and “color/symbol combination.” For example, “3-4-2” as the numeral combination indicated that the target drug was located on monitor-3, fourth row from the top, and second column from the left. Likewise, “➂ ” as the color/symbol combination indicated that the drug was located on monitor-3, on the blue line, and second from the left. For “color/symbol combination,” five colors (red, yellow, green, blue, and black) were used as the location display, and a colored line was shown above each row in the drug rack. The five target drugs were not located in the same row and column (x–1–1, x–2–2, x–3–3, x–4–4, or x–5–5; where “x” indicates the monitor number). Details of the dispensing information are presented in Table 1. In a series of studies as dispensing verifications, we generated 14 prescription slides and 42 drug rack slides to perform a total of seven pairs of dispensing verifications. The order of verifications was random. Notably, we only analyzed the data acquired from five of the seven pairs of verifications in this study.

Table 1 List of verification informationVerification procedure

An outline of the verification task using the eye-tracking method is shown in Fig. 1. We connected five notebook computers to 27-inch monitors (monitor-1, -2, -3, -4, and -5) to operate the slides. The drug rack area (length 34 cm × width 200 cm) on monitor-1, -2, and -3 was on the upper stage, and the prescription area (length 34 cm × width 60 cm) on monitor-5 directly below monitor-2 was on the lower stage. Monitor-4 for the prescription inquiry was arranged to the left of monitor-5. A pharmacist wearing an eye tracker was seated on a chair 100 cm from monitor-5 to read the prescription slide. By showing the drug rack area (monitor-1, -2, and -3) and the prescription area (monitor-5) simultaneously, we could investigate the gaze movements of pharmacists during the dispensing process. Using Tobii Pro Lab Analyzer with the recorded motion video, we could also assess several categories, such as gazing point (center point in the circle), gazing time (size of the circle), and gaze movement (line between center points of circles).

Fig. 1

Outline of the verification process using the eye-tracking method. Gaze movements obtained by eye-tracking and analyzed using Tobii Pro Lab Analyzer were mainly classified into the two categories fixation (stagnation for a certain time) and saccade (quick movements of the eyeballs). We analyzed a series of dispensing processes by showing the prescription (length 34 cm × width 60 cm) and drug rack (length 34 cm × width 200 cm) areas. The red dotted line in the figure represents the boundary between the two areas, the left side from the center oblique line on monitor-2 corresponds to the drug rack example using the display method of “numeral combination,” while the right side from the line corresponds to the “color/symbol combination.” The center point in the circle, the size of the circle, and the line between the center points of circles indicate the gazing point, gazing time, and gaze movement, respectively. A pharmacist wearing an eye tracker was seated on a chair 100 cm from monitor-5 and performed several pairs of dispensing verifications in random order

To ensure the accuracy of the eye tracker, we performed gaze calibration with each pharmacist before conducting the verification experiments. To get used to the verification process, pharmacists were allowed to practice with several training slides in advance. Smooth dispensing as usual was prioritized in this verification task, and if a pharmacist noticed a mistake in the dispensing process, they were allowed to correct the mistake immediately. Furthermore, if the pharmacist determined that there was an issue with the prescription content, they would point at the monitor-4 for further inquiry regarding the prescription. We analyzed a series of verification processes, from confirming the dispensing information in the prescription area to pinpointing the five target spots in the drug rack area. The main steps for dispensing verifications were as follows:

1)

A pharmacist gazed at a given position.

2)

An assistant switched to a prescription slide and three drug rack slides simultaneously when the “Next” signal was indicated by the pharmacist.

3)

The pharmacist read out “total amount” of a target drug while pinpointing the target spot and repeated this process a total of five times.

4)

The assistant switched to a rest slide when the “Next” signal was indicated by the pharmacist.

5)

The verifications, using 14 prescription and 42 drug rack slides, were repeated with voluntary breaks.

Definition of the five paired models

First, we set up two pairs of models (A1-A2 and B1-B2; Fig. 2) to compare the difference in gaze movements between the left and right areas. The location display of the five target drugs in models A1-A2 was expressed by a numeral combination, whereas that in models B1-B2 was expressed by a color/symbol combination. The five target drugs in each pair of models were the same, but their locations differed in that they were arranged left-right symmetrically regarding the center line.

Fig. 2

Arrangement of the five target drugs in two pairs of models A1-A2 (upper side) and B1-B2 (lower side). The location displays of the five target drugs in the two pairs of models A1-A2 and B1-B2 are indicated using the “numeral combination” and “color/symbol combination” methods, respectively. These five target drugs in each pair of models (A1-A2 and B1-B2) are the same, but their locations are arranged left-right symmetrically regarding the center line (left vs right area). The location spots of the five target drugs in models A1-A2 and B1-B2 are shown as white circles (○) and white triangles (▽), respectively

Second, we set up three pairs of models (C1-C2, D1-D2, and E1-E2; Fig. 3) to compare the difference in gaze movements between the location displays in “numeral combination” and “color/symbol combination” in the left, center, and right areas. The five target drugs and their locations were the same in each pair of models, but their location display methods differed in being “numeral” or “color/symbol” combinations. A summary of the two pairs (A1-A2 and B1-B2; Fig. 2) and three pairs (C1-C2, D1-D2, and E1-E2; Fig. 3) of models are given below.

Model A1-A2: comparison between the left and right areas by the location display using “numeral combination.”

Model B1-B2: comparison between the left and right areas by the location display using “color/symbol combination.”

Model C1-C2: comparison between “numeral combination” and “color/symbol combination” in the left area.

Model D1-D2: comparison between “numeral combination” and “color/symbol combination” in the center area.

Model E1-E2: comparison between “numeral combination” and “color/symbol combination” in the right area.

Fig. 3

Arrangement of the five target drugs in three pairs of models C1-C2 (left area), D1-D2 (center area), and E1-E2 (right area). The location displays of the five target drugs in the three pairs of models C1-C2, D1-D2, and E1-E2 are indicated using the “numeral combination” (upper side) and “color/symbol combination” (lower side) method. These five target drugs and their locations in each pair of models (C1-C2, D1-D2, and E1-E2) are the same. The location spots of the five target drugs in models C1-C2, D1-D2, and E1-E2 are shown as black circles (●), triangles (▼), and squares (■), respectively

Verification items and classifications

To dispense a target drug in the verification task accurately, a pharmacist needs to visually recognize the four items (a) drug name, (b) drug usage, (c) location display, and (d) total amount in the prescription area and specify exactly a target spot in the drug rack area. Furthermore, the pharmacist also needs to move the visual line up and down as (e) vertical movement between two areas. Here, a total of gaze points in the prescription area, those in the drug rack area, a total of vertical movements passing through a boundary between two areas, and the time required to dispense the five target drugs were measured in the four classifications Gaze 1, Gaze 2, Passage, and Time. We calculated the differences in gaze movements between the left and right areas according to each location display of “numeral combination” or “color/symbol combination” (models A1-A2 and B1-B2, respectively), as well as those between the location displays of “numeral combination” and “color/symbol combination” in the left, center, and right areas (models C1-C2, D1-D2, and E1-E2, respectively).

Gaze 1: A total of gaze points in four items of (a), (b), (c), and (d) in the prescription area.

Gaze 2: A total of gaze points including five target spots in the drug rack area.

Passage: A total of (e) vertical movements between the prescription and the drug rack areas.

Time: A length of time required to dispense five target drugs.

Calculation of reconfirmation frequency per target drug in three areas (left, center, and right)

To analyze the differences in dispensing complexities between the display methods “numeral combination” and “color/symbol combination” in three areas (left, center, and right), it is necessary to compare the reconfirmation frequency of items (a)–(e) per target drug between each pair of models C1-C2, D1-D2, and E1-E2.

Thus, we first defined the “essential number” as the minimum checks of items (a)–(e) required for dispensing a target drug as follows: (a’) 2 points, (b’) 3 points, (c’) 1 or 2 points, (d’) 1 point, and (e’) 2 times. Concerning the display method “color/symbol combination,” the number of gaze positions varied depending on the drug location. For instance, the essential number of the (c) location display was one in the left area but two in both the center and right areas. Using these definitions, we then calculated the average reconfirmation frequencies (a”)–(e”) by subtracting the essential numbers (a’)–(e’) from the average gaze values (a)–(e) per target drug. Following this new definition, the average reconfirmation frequencies (a”)–(e”) per target drug were calculated for the three pairs of models.

Data analysis

Using the gaze category (fixation, saccade) data, we analyzed the gaze frequency in the prescription area, that in the drug rack area, the number of vertical movements between the prescription and drug rack areas, and the length of the dispensing time. Data were presented as the mean ± standard deviation of participants, and differences were analyzed using the paired t-test. P value of <0.05 was considered statistically significant; P values of <0.01 and <0.001 were considered highly significant. The paired t-test was performed using JMP Pro 15 statistical software.