Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
This project has received funding from the European Union's Horizon 2020 Research and Innovation Programmeunder Grant Agreement no 730998
Laboratoire MESuRS “Modélisation, Epidémiologie et Surveillance des RisquesSanitaires” - Cnam
QWL & psychosocial risks:How to identify and prioritize
psychosocial risk factorsimpacting stress level
Mounia N. HOCINEWilliam DAB
Patrick LEGERONGilbert SAPORTA
www.inclusivegrowth.eu 2
Outline
Introduction
Data
Statistical methodology
Resultats
Current work: alternative approaches
Conclusion
www.inclusivegrowth.eu
Work-related stress
• A major public health issue
has negative effects on physical and psychological health
is an inevitable part of organizational life
4
The aim: Reduce work-related stress level
Decision makers would like to be provided with
statistical tools that can help them identify risk factors
requiring a priority action
www.inclusivegrowth.eu
Quantitative assessment
• Theoretical frameworks have been successful in generating and collecting data on work-related stressors: – Demand/Control Imbalance Model
– Effort/Reward Imbalance Model
M. Gollac recommendation: setting up a mental health monitoring system
5
www.inclusivegrowth.eu
Motivating data
A system of data collection on stress and stressorsduring the yearly compulsory routine visit with the company’s occupational physician
Measurement tools 2 anonymous online questionnaires (response rate > 80%)
Database: 10 000 anonymous employees randomly drawn from Stimulus data collected over 2011-2012.
6
www.inclusivegrowth.eu
Stress measurement tool
• 1st questionnaire: 25 items to measure individual psychological stress at work 8-point Likert scale.
• Example:
“I'm confused and I lack focus and concentration” , answer varies from 1 “not at all” to 8 “enormously”
Build a stress score = responses (25-200)
7
www.inclusivegrowth.eu 8
• 2nd questionnaire: 58 items to measure the impact of job characteristics (stressors) 6-point Likert scale.
Example 1: “My company does not care about employees well-being” (negative) 0 “totally disagree” to 5 “totally agree”
Example 2: “I know clearly what I am expected to do at work” (positive) 0 “totally agree” to 5 “totally disagree“
Stressors measurement tool
www.inclusivegrowth.eu 9
• The 58 stressors are grouped into five blocks:
- Work Context (14)- Job control (14)- Relationships at workplace (12)- Performed tasks (12)- Recognition (6)
Stressors blocks
www.inclusivegrowth.eu
Cooper index
• Developped by Cooper & Clarke, Health, Risk & Society 2000
• Provides companies a quantitative risk assessment approach to prioritize psychosocial risks at work.
• Approach: identify stressors related to high stress level:
Risk level = exposure x consequences
10
Stressor level Cor² (stressor; stress)
www.inclusivegrowth.eu
Advantage & Limits
1) Easy to use but questionnable! r² x m is not a risk(or an impact) measure
2) Strong hypothesis: Acting on a given stressor does not impact the other stressors ! No consideration of inter-correlations between stressors
12
www.inclusivegrowth.eu
Suggested approach (1st step)
• Partial-Least Square-Structural Equation modeling that allows:– To investigate the impact of job characteristics on
perceived work-related stress in a multidimensional way
– To predict blocks of stressors requiring priority action from managers in preventing high level of work-related stress.
Stressors are correlated and grouped into a block structure generated from a conceptual model
14
www.inclusivegrowth.eu 16
Latent variables No. of Items Cronbach a
1st questionnaire: « stress »
Work-related stress 25 0.94
2nd questionnaire: « stressors »
Work context 14 0.88
Job control 14 0.80
Relationship 12 0.87
Tasks 12 0.62
Recognition 6 0.81
Homogeneity of the blocks
www.inclusivegrowth.eu 17
Validity of the blocks
LV MV Outer weight 95% Confidence Interval
Work context
Context_01 0,116 0,110 0,122
Context_08 -0,069 -0,074 -0,064
Context_15 -0,077 -0,082 -0,072
Context_21 -0,096 -0,100 -0,092
Context_23 -0,096 -0,100 -0,092
Context_30 -0,088 -0,092 -0,084
Context_32 -0,085 -0,089 -0,081
Context_34 0,090 0,086 0,095
Context_40 -0,080 -0,084 -0,076
Context_41 0,096 0,090 0,102
Context_43 -0,080 -0,085 -0,076
Context_52 0,092 0,087 0,097
All outer weights are statistically significant (a = 5%)
www.inclusivegrowth.eu
Predict stress: PLS regression
18
Context Control Recognition Relationship Tasks Stress
Context 1.00
Control 0.78 1.00
Recongnition 0.72 0.63 1.00
Relationship 0.69 0.67 0.60 1.00
Tasks 0.61 0.72 0.54 0.53 1.00
Stress -0.52 -0.63 -0.43 -0.51 -0.52 1.00
• Correlation between latent variables:
• A high multi collinearity between the 5 stressorsblocks PLS regression
www.inclusivegrowth.eu
Path coefficients
19
0.15
Path coef.
(R² = 40%)
0.18
0.12
0.14
0.15
efficients cannot be used directly to provide
decision makers with ranked predictors
www.inclusivegrowth.eu 20
Importance-Performance Analysis
• A graphical grid for decision making (Martilla & James, The Journal of Marketing 1977)
www.inclusivegrowth.eu 21
• Performance: measured as the score mean of 10 000
responses
• Importance: calculated using the suggested formulae:
Importance (kth item) = |Outer weight (kth item in jth block)| x
Path coefficient (jth block, stress)
Importance-Performance Analysis
www.inclusivegrowth.eu
Stressors to maintain
Task_05: « I frequently see the work pile up without being able to eliminate the backlog »
Task_07: inverse of «My work gives me many opportunities to perform interesting tasks »
Task_01: inverse of «My work has meaning to me »
Task_02: «My job involves monotonous and repetitive tasks »
Contro_03: inverse of « I can achieve professional life—personal life balance »
Contro_14: « I'm undergoing or I expect to undergo an undesirable change that might affect my career »
23
www.inclusivegrowth.eu
Priority action
1. Task_11: « I have to work fast in a short timeframe »
2. Recon_01: «My promotion prospects are weak »
3. Recon_02: inverse of «My company offers me interesting career opportunities »
4. Task_09: « I work in a noisy and hectic atmosphere »
5. Recon_04 : inverse of « I am rewarded when I reach my goals »
24
www.inclusivegrowth.eu
Limitations
1) The validity of the underlying conceptual model is questionable
2) The previous method SEM-path modeling combined with IPA is not easy to use by decision makers
3) Confusion between correlation and causality!
acting on a stressor would cause changes in other stressors
25
www.inclusivegrowth.eu
1. Weifila “Weighted first last”
A variance decomposition method in linear regression context (Y=sress score)
𝐸(𝑌) = 𝛼 + 𝛽1𝑋1 +⋯+ 𝛽𝑝𝑋𝑝
Predictor’s importance 𝑊(𝑗) = weighted average between:
• “First” allocation: 𝐹𝑖𝑟𝑠𝑡(𝑗) = 𝑐𝑜𝑣(𝑦1; 𝑋𝑗)
• “Last” allocation: 𝐿𝑎𝑠𝑡(𝑗) = 𝑠𝑟²(𝑗) « semi-partial R² »
27
𝑊 𝑗 = 𝐿𝑎𝑠𝑡 𝑗𝐹−𝑅2
𝐹−𝐿+ 𝐹𝑖𝑟𝑠𝑡 𝑗
𝑅2−𝐿
𝐹−𝐿𝑖𝑓 𝐿 < 𝑅2 < 𝐹
𝐿𝑎𝑠𝑡 𝑗𝑅2−𝐹
𝐿−𝐹+ 𝐹𝑖𝑟𝑠𝑡 𝑗
𝐿−𝑅2
𝐿−𝐹𝑖𝑓 𝐹 < 𝑅2 < 𝐿
www.inclusivegrowth.eu
• A combination of tree predictors of Y=stress score
• Each tree depends on the values of a random vector sampledindependently with the same distribution for all trees in the forest
• Predictor’s importance = the frequency of the item occurrence in the forest trees
• We used “Gini” a local criteria based on the heterogeneity decrease
The importance of an item is a weighted sum of the induced heterogeneity
28
2. Random Forest
www.inclusivegrowth.eu 29
Consider Y= stress score
The association between Y and the predictors 𝑥𝑗 𝑗 = 1,… , 58
can be estimated using a linear regression
To predict Y1 = 1 if Y>110, 0 if not
Logit (p) = log 𝑝(𝑌1=1)
𝑝(𝑌1=0)= 𝑏0 + 𝑏1𝑥1 +⋯+ 𝑏𝑝𝑥𝑝
Consider now the average exposure E(𝑥𝑗) to the predictor 𝑥𝑗
EF 𝑥𝑗 =𝑒𝑥𝑝 𝑏𝑗 −1 ∗𝐸 𝑥𝑗
1+ 𝑒𝑥𝑝 𝑏𝑗 −1 ∗𝐸 𝑥𝑗“An epidemiological impact measure”
3. Etiologic fraction (EF)
www.inclusivegrowth.eu 30
Weifila AR R. Forest
PSRF09 PSRF09 PSRF09
PSRF54 PSRF13 PSRF25
PSRF25 PSRF54 PSRF54
PSRF14 PSRF24 PSRF45
PSRF13 PSRF41 PSRF13
PSRF44 PSRF25 PSRF03
PSRF09: inverse of “I can manage my professional life and personal life”PSRF54: “I am living/expecting an undesirable change likely to affect my career”PSRF25: “I often see the work pile up without being able to absorb the delay”
Priority action: comparison
www.inclusivegrowth.eu
Conclusion
• The four metrics lead to similar results
• For organizational strategies to prevent and manage stress, decision-makers should act in priority on the level of the 5 “stressors to improve”
• Etiological fraction-based approach is a relevant-easy to implement tool to help decision making
31
www.inclusivegrowth.eu
NEW DATASET
QWL barometer survey: “ Baromètre Travail et Santé Psychique”
• A cross-sectional study: 3200 assets
• A representative sample of the French population according to: age, gender, occupation
• Survey conducted by Ipsos between Fabruary & March 2018
32
www.inclusivegrowth.eu
Sociodemographic profile of the interviewed employees
51% 49%
9
23
27
37
4
25-34 y
35-44 y
60 y or +
18-24 y
21
22
22
11
24
Ile de France Nord OuestNord Est Sud OuestSud Est
Region
# children 18 years old or less
52
22
20
5
1
Aucun
1
2
3
4 et +
45-59 y
- 35 y. old: 32%
+ 35 y. old: 68%
NEW DATASET
www.inclusivegrowth.eu
Factorial analysis
Somatisation
GHQ-28
34
Désadaptation
sociale
Dépression
sévère
Anxiété et
insomnie
General Health Questionnaire
Unidimentionnality of the GHQ-28
www.inclusivegrowth.eu
PSRF1
36
PSRF2
….
PSRF43
Score GHQ
PSRF44
44 PRF documented
Psychosocial Risk Factors
www.inclusivegrowth.eu
Confounding factors
Gender, age
Tobacco, alcohol
Physical health
Sick leave for psychic problem
Drug consumption
Medical check-up
Week-end work, night work
Sport, income, …etc.
37
www.inclusivegrowth.eu
Linear Regression ModelItems Coefficient
PSRF9 1,96**
PSRF18 1,67**
PSRF5 1,64**
PSRF8 1,24**
PSRF43 0,85*
PSRF30 0,81*
Priority action
PSRF9 1) Inverse of “I can manage my professional life and my personal life”
PSRF18 2) Communication and information exchange within my company are
satisfactory
PSRF5 3) Inverse of “I know I can rely on people I work with”
PSRF8 4) I have to deal with a lot of complex and numerous information
PSRF43 5) I’m worried for my professional future
* < .01
** < .001
www.inclusivegrowth.eu
Conclusion & Perspectives
• Similar main risk factor identified (external validation)
• Causal inference analysis is needed to determine stressors on which to act in order to reduce psychosocial disorders associated with stress.
Follow-up data: individual trajectories on stress and psychosocial risk factors needed
39
www.inclusivegrowth.eu
Work conditions and infectious risk
2013-2016: MOTILIS « Modélisation de l’Organisation du Travail
et de son Impact sur les infections Liées aux Soins »
Impact of organizational factors on stress and fatigue in
French intensive-care healthcare workers
Jones, Hocine et al. Inter J Nursing Studies 2015
40
www.inclusivegrowth.eu
Work conditions and infectious risk
41
2017-2020: STRIPPS research project
« modelling work organization and its impact on
healthcare-associated infections »
Partner: CRTD, Cnam
Prospective cohort study
To propose optimal organizational strategies for controlling
infectious risk.
www.inclusivegrowth.eu
Data collection
Longitudinal data collection over 1 year in 5 hospitals in
Paris (35-40 randomly selected wards)
Data collection started in February 2018
4 time points: T0, T0+4 months, T0+8 months, T0+12
months
www.inclusivegrowth.eu
Methodology
Quantitative/qualitative approach
Quantitative: Administered questionnaires (~750 HCWs),administrative databases, hand hygiene compliance audits,…
Longitudinal data analyses, hierarchical models, causalityanalyses
Qualitative: interviews with HCWs and days of observationby the ergonomist
Analysis of activity
www.inclusivegrowth.eu
Concluding sheme
45
Conceptual model built by experts
(P.Légeron)
Measurement tools: questionnaires on stress & psychosocial risk factors(validated instruments)
Data collected during a routine visit with the
occupational GP (high quality)
PLS Structural Equation Model: predicted successfully stress
(deal with colinearity)
Importance-performance analysis: prioritized stressors
regarding their impact on stress (graphical tool)
Step 1
Step 2
Causality analysis (Priority action)
Step 3
www.inclusivegrowth.eu
Some references
1. Hocine MN, Aït Bouziad K, Légeron P, Dab W, Saporta G.How to Identify and Prioritize Psychosocial Factors Impacting Stress Level. PLoS One. 2016 Jun 15 ;11(6):e0157078.
2. Henri Wallard. Using Explained Variance Allocation to Analyse Importance of Predictors. 16th ASMDA Conference Proceedings, 30 June – 4 July 2015, Piraeus, Greece
3. Clarke, S. G., & Cooper, C. L. The risk management of occupational stress. Health, Risk & Society 2000.
4. Vinzi, E, Russolillo, G. Partial least squares algorithms and methods. WileyInterdisciplinary Reviews: Computational Statistics 2013.
5. Martilla, J. A., & James, J. C. Importance-performance analysis. The journal of marketing 1977.
6. Bühlmann, P. Causal statistical inference in high dimensions. Mathematical Methods of Operations Research 2013.
46