Download - Predicting Medical Professionals’ Wellbeing Using Wearable ......3Department of Electrical and Computer Engineering, Rice University, Houston, Texas, U.S.A. Wellbeing in Medical

Transcript
Page 1: Predicting Medical Professionals’ Wellbeing Using Wearable ......3Department of Electrical and Computer Engineering, Rice University, Houston, Texas, U.S.A. Wellbeing in Medical

Predicted0: Unhappy 1: Happy

Actual0: Unhappy 41 3651: Happy 134 3360

• Fitbit data

• Wellbeing Indexes every morning and evening

30 days

[1] Natasha Jaques, Sara Taylor, Asaph Azaria, Asma Ghandeharioun, Akane Sano and Rosalind Picard. Predicting students’ happiness from physiology, phone, mobility, and behavioral data. In 2015 International Conference on Affective Computing and Intelligent Interaction, September 2017.[2] Andrew J. K. Phillips, William M. Clerx, Conor S. O’Brien, Akane Sano, Laura K. Barger, Rosalind W. Picard, Steven W. Lockley, Elizabeth B. Klerman & Charles A. Czeisler. Irregular sleep/wake patterns are associated with poorer academic performance and delayed circadian and sleep/wake timing. Nature, Scientific Reports 7, June 2017.[3] Akane Sano, Sara Taylor, Andrew W McHill, Andrew JK Phillips, Laura K Barger, Elizabeth Klerman, Rosalind Picard. Identifying Objective Physiological Markers and Modifiable Behaviors for Self-Reported Stress and Mental Health Status Using Wearable Sensors and Mobile Phones: Observational Study. 2018.

• Analyze pilot study data of medical professionals to understand relationships between wearable data and their wellbeing

• Compare the findings to the previous research about college students

• Build machine learning models to predict their happinessUltimately we aim to develop technologies to support them (e.g. optimize their work shift schedule, send notifications (ex. ”you should take a rest now” ”if you sleep x hours from y pm today, you will be able to be more alert tomorrow at work”)

Predicting Medical Professionals’ Wellbeing Using Wearable and Survey Data Mei Fukuda1,2, Cheng Wan3, Han Yu3, Akane Sano3

1Department of Mathematical and Computing Science, Tokyo Institute of Technology, Tokyo, Japan2TOMODACHI STEM Program 2019, Rice University, Houston, Texas, U.S.A.

3Department of Electrical and Computer Engineering, Rice University, Houston, Texas, U.S.A.

Wellbeing in Medical Professionals

Data Analysis to Detect Relationship between Wearable Data and Wellbeing Index

Machine Learning to Predict Wellbeing

Wearable and Survey Data

Methods

Acknowledgements

Conclusion

References

Medical professionals work long hours with irregular shift schedule. Moreover, their work especially requires accuracy. Because of this, they often suffer from fatigue and depression.Our project aims to analyze medical professionals’ data to understand

the relationships among their work schedule, sleep and wellbeing ultimately to make them more efficient and happier. This poster introduces some preliminary data analysis and modeling results using our pilot experiment data.

Goals

1 2 3

4 5 6 7 8 9 10

11 12 13 14 15 16 17

18 19 20 21 22 23 24

25 26 27 28 29 30

5 medical professionals in Japan

sleepy alertsad happysluggish energeticsick healthystressed out calm relaxed

1000

• Survey responses

• Work time shift schedule (0: no duty, 1: 8:30-16:30, 2: 16:30-0:30, 3: 0:30-8:30)

Morning Questions

I did not get any sleep in the last 24 hours.How many times did you take a nap yesterday?What time did you go to sleep last night?How many times did you wake up last night?How did you wake up this morning?What time do you start work today?

Evening Questions

Did you attend any other activity besides work? How many total servings of caffeine did you have today? When was the latest hour you consumed caffeine? Besides caffeine, did you consume any other medication, drugs or alcohol today? Did you feel any fear or anger today?

Mon

Tue

Wed

Thu

Fri

Sat

Sun

Social Jetlag

Workdays’Sleep Midpoint *1

Non-workdays’Sleep Midpoint

0:00 7:00

0

0

0

0

1

:

1

1

0

0

0

0

1

:

1

0

0:11

0:12

0:13

0:14

0:15

:

7:30

7:31

!"##$ %#&'"()*+, -./#0 = 200 ×∫6789

: ; : ; + = /+

> − τ− 0.5

+24hours

↑↑

We would like to thank our collaborators at Mie University (Drs. Shimaoka, Kawamoto and Ito).This research project was conducted as part of 2019 TOMODACHI STEM @ Rice University Program funded by

U.S.-Japan Council’s TOMDACHI Initiative. I would like to thank the members of the Prof. Akane Lab for their research mentorship. I also would like to thank TOMODACHI STEM program faculty and staff, Prof. Junichiro Kono, Sarah Philips, Aki Shimada and my fellow participants for their support and encouragement. For more information, visit http://tomodachistem.rice.edu/. Contact: Mei Fukuda ([email protected])

1. Calculate correlations between wearable featuresand wellbeing indexes based on each hypothesis ex.) Longer sleep duration → Happier

2. Apply result of 1. to build machine learning models

0

0

1

1

1

:

1

0

C#(.: E = F

G∑IJFG 0I ,

!M !+(./()/ M#N*(+*O. : P = F

G8F∑IJFG (0I − E)S

• Basic Statistics:

• Sleep Duration

• Sleep Regularity Index

• Social Jetlag Index

*1 !"##$ C*/$O*.+= !"##$ !+()+*.& >*T#+ (!"##$ M')(+*O.)/2

Wellbeing Index P-Value *Coefficient

HappinessMorning 0.001 *0.032Evening 0.382

HealthinessMorning 0.002 *0.032Evening 0.659

CalmnessMorning 0.005 *0.028Evening 0.864

: : :

Morning Indexes:Statistically significant with significance level of 0.05

WellbeingIndex

Happiness Healthiness Calmness …

Mean SD Mean SD Mean SD …

P-Value 0.582 0.418 0.943 0.966 0.561 0.227 …

• There’s a possibility to predict medical professionals’ wellbeing by analyzing wearable data

• We may be able to compute relationships more accurately If we also collect more various data similarly in our previous college student research[1]

• Use other ways to analyze the data and create a machine learning model

• Try another definition of ”Happy day and Unhappy day” ex.) 0-30: unhappy, 30-70: do not include, 70-100: happy

• Try other indexes to create machine learning model

WellbeingIndex

Happiness Healthiness Calmness …

Mean SD Mean SD Mean SD …

P-Value 0.755 0.763 0.937 0.656 0.176 0.287 …

�Figure 1. scatter graph for happiness of morning and sleep duration

�Table 1. p-value of wellbeing indexes and sleep duration

�Table 2. p-value of wellbeing indexes and social jetlag index

�Table 3. p-value of wellbeing indexes and sleep regularity index

�Figure 2. social jetlag index

�Figure 3. sleep regularity index

• Heartrate, Steps and Calories

Future Research

• Work Time Shift

Features Happiness Healthiness Calmness …

HeartrateMean 0.060 0.081 0.041 *-0.635 …SD 0.357 0.038 *1.461 0.475 …

Steps 0.809 0.373 0.458 …Calories 0.341 0.015 *-0.010 0.032 *-0.009 …

�Table 4. p-value of Healthiness and features

Shift Happiness Healthiness Calmness Energetic Alertness1: 8:30~16:30 0.404 0.316 0.586 0.213 0.6292: 16:30~0:30 0.390 0.431 0.516 0.162 0.6543: 0:30~8:30 0.320 0.366 0.106 0.062 0.034 *-13.617Time [hours] 0.494 0.412 0.998 0.260 0.199

�Table 5. p-value of wellbeing indexes and shift

� Figure 4.scatter graph forHealthinessand SD of heartrate

• Linear Mixed Effects Model:

Used for regression analysis with dependent data

Significance Level, P-ValueOutput: Coefficient, Standard Error,

t:timeT:daystocalculateregularity

e + =f0 if awake at +1 if sleep at +

τ : 24hours

[minutes]

Hap

pine

ss o

f Mor

ning

Sleep Time in Bed

Hea

lthi

ness

of M

orni

ng

SD of Heartrate

� Features:

� Feature Normalization:

�Model:

� Test and Train Sets

� Results

n − Eσ

SVM (Support Vector Machines)a set of related supervised learning methods used for classification and regression+ applied RBF (Radial Basis Function) as Kernel

• Sleep Duration• Heartrate Mean• Heartrate SD

• Steps• Calories• Work Time Shift

� Label: Happiness of Morning (1: ≥50, 0: <50)

Mean → 0Variance → 1

� Test: one of subjects Test: 30% data

Predicted0: Unhappy 1: Happy

Actual0: Unhappy 25 381: Happy 30 111

Train: rest of dataIteration: 100 times

Train: rest of subjectsIteration: 5 times

�Table 6.,7. F1 score and a total of confusion matrix for classification A

�Table 8.,9. F1 score and a total of confusion matrix for classification B

(Sleep duration, Sleep starting time, Heartrate, Steps, Calories)

F1 ScoreMean 0.923SD 0.0492

F1 ScoreMean 0.930SD 0.0292

*Coefficient

*Coefficient