Predictive value of Heart Rate Variability measurements and the Brief Resilience Scale for workability and vitality

2.1Study design

A prospective observational follow-up study was carried out with baseline measurements between November 2015 and November 2016. Follow-up period was between 24 and 48 months.

2.2Setting

Because the study aimed for applicable results within a real-life context, data were collected during routine Workers’ Health Assessments (WHA) conducted on regular clients of an Occupational Healthcare Supplier (OHS) servicing 18 moderate and large companies in a variety of sectors (education, transport, automotive industry, housing association, water board, food processing, waste processing, metal industry, retail, finance) in the Netherlands. The WHAs were performed by the independent OHS to promote sustained employability, and were used to identify specific work-related health risks in order to prevent injuries or disease through intervention [17]. All workers from participating companies were invited. Participation was voluntary, and companies were responsible for the invitation process. Legislation in the Netherlands requires companies to offer employees a WHA at least every four years. Therefore, our maximum follow-up period was four years. The minimum follow-up period was two years, which was sufficient to allow health changes to occur.

2.3Ethical standards

The Ethics Board at the University Medical Center Groningen in the Netherlands confirmed (August 20, 2015) that formal approval of the study was not necessary because all workers were only subjected to routine care. The study was performed in accordance with the Declaration of Helsinki of 1964 and its later amendments. Participants signed informed consent forms stating that their anonymized data could be used for scientific purposes to improve the quality of the WHA service. This paper adhered to the EQUATOR network STROBE guidelines for observational cohort studies.

2.4Participants

All workers (from participating companies described in the Setting section) between 18 and 65 years old were included. Workers absent from work at the time of the measurements did not participate. Pregnant women were excluded. Workers with pacemakers or beta-blockers were excluded because of the influence on HRV.

2.5Measurements

2.6Data handling

Questionnaire data were entered in the OHS company software by the worker and scores were processed automatically according to each questionnaires guidelines, except for the BRS-DLV. Patient characteristic and HRV biometry results were entered in the same software manually by the OHS. All data were then compiled into an Excel file and de-identified by the OHS. The data were then sent to the researchers.

The researchers processed the data of the BRS-DLV, which is constructed with three reverse phrased questions, for example “I tend to take a long time to get over set-backs in my life” to enable detection of an inconsistent pattern. These answers were reversed in SPSS during data-processing. In line with the validation study of the BRS-DLV [23], the researchers manually removed inconsistent cases that indicated respondents either did not understand or read the responses. Only cases with complete data for WAI-T1, UWES-9-Vitality-T1, HRV-T0, BRS-T0, age, gender and correct follow-up period were used for statistical analysis (T0 = baseline, T1 = follow-up). Missing data are presented in the result section.

MHRR data were categorized according to data handling description in a previous study [22]: “..the data were divided into subgroups because literature indicates low HRV is related to health problems [11, 13, 15]. In absence of validated meaningful cut-off values, statistical criteria were used to determine cut-off values. Because only low scores are known to have a clinical value [11, 13, 15], cut-off values are provided on the left side of the normal distribution of MHRR of a healthy reference group, corrected for age (proprietary; Emwave Pro Plus® software): very low MHRR when MHRR > 2 SD below mean, low HRV when MHRR is 1-2 SD below mean, (above) average HRV when the deviation is higher than 1 SD below mean.”

2.7Statistical analysis

Descriptive statistics were calculated for all variables. A large number of workers (n = 1,196) were excluded due to missing data or a follow-up period < two years. To test for selection bias, differences between the included and excluded group were compared for demographic, personal and health variables. A t-test with Bias-corrected and accelerated (BCa) bootstrap procedure (1,000 samples) was used for continuous data for continuous data (P < .05). A Kruskal-Wallis H test was used for ordinal and nominal data. Difference between the HRV groups was tested with a Kruskal-Wallis H test, followed by a pairwise comparison.

To assess the joint predictive ability of HRV, BRS-DLV and the covariates (BMI, gender and age) on workability and vitality, yet eliminating the least important variables, a backward stepwise multiple regression analysis (p < .05) was used. The 95% Confidence Interval around B is calculated with a BCa bootstrap procedure (1,000 samples) to overcome bias from heteroscedasticity and normality of residuals. With the backward procedure, the least significant variable was excluded from the model step by step until all remaining variables contributed significantly (p < .05) and collinearity was acceptable (Variance Inflation Factor ≈ 1). The backward stepwise multiple regression procedure is preferred to a forward procedure, because all variables are taken into account simultaneously and multi-collinearity is accounted for. Sensitivity analyses for HRV was performed by entering raw MHRR data in the regression analysis. IBM-SPSS (version 26) software was used for statistical analysis.

3.1Participants

At baseline, 1,624 workers employed at 18 companies participated; 630 were measured before the follow-up period of 24 months and were therefore excluded. Another 566 cases were excluded because they missed (a combination of) variables (143 workers missed the UBES-9-vitality-T1; 75 the WAI-T1; 191 the MHRR-T0; 140 the BRS-score-T0). Six workers were excluded because the BRS data were inconsistent. This led to 428 included workers employed at four companies for the final modeling. WAI and UBES baseline score of the included and excluded group were compared to rule out selection bias; the groups did not differ significantly (p < .01). The majority (n = 402) worked at a holding company in the automotive industry. The others (n = 26) were employed at a housing association, meat processing plant or water board company.

3.2Personal, work and health characteristics at baseline

Personal and work characteristics at baseline are presented in Table 1. Mean age was 41 years (SD 9.5); the majority of participants were male. Most workers had a fixed contract, did not work in shifts and level of education was intermediate or high. Median working years was 7.5 years with a right-skewed distribution, and median working hours a week was 37.6 hours with a left skewed distribution. The included group was significantly higher educated, worked more in mentally demanding work and less in shifts than the excluded group (p < 0.05). There were more females in the included group. Other differences were minor and not relevant or significant. Results of both included and excluded participants are presented in Appendix B.

Health characteristics at baseline are shown in Table 2. Median BRS score was 3.6 (range 2.0–5.0) and median MHRR was 22.0 beats/minute (range 2.8 – 70.0 beats/minute). The Kruskall-Wallis H test showed that groups differ significantly (p < .001). However, the pairwise comparison showed that the very low and low MHRR group did not differ (p = .45). The (above) average group was large, as a result of the categorization procedure. No relevant differences in health characteristics at baseline were found between the included and excluded workers (Appendix C). Univariate partial correlation between BRS and MHRR, controlled for age, was.04 (BCa 95% interval: –.05 to.13).

3.3Dependent variables at follow-up

Median duration between T0 and T1 was 38.4 months (minimum 24–maximum 46 months). With a median WAI-score of 42.7, workability at follow-up was good; the median UWES-9-vitality score was 4.2, indicating a moderate vitality (Table 3).

3.4Regression analysis

Table 4 shows the final models for WAI-score and for UWES-9-vitality score. MHRR was excluded in the second step, after gender, for the Workability model and in the first step for the UWES-9-vitality model. All steps of the linear backward regression procedure can be found in Appendix D. The variance inflation factor (VIF) values were below 1.1 in all steps and 1.0 in the final models, indicating there was no collinearity between variables. The sensitivity analyses with raw MHRR data (results not shown in detail) revealed a similar picture. MHRR in the very low and low categories did not differ significantly; therefore we additionally tested the effect of merging these groups, which did not change the outcome (results not shown in detail). MHRR did not contribute significantly to prediction of Workability or UWES-9-vitality. In the final models, the contribution of resilience, measured using the BRS-DLV, was modest but significant for the prediction of WAI-score (R² = 9.2%) and UWES-9-vitalty score (R² = 7.3%). Age was included as a significant co-variate only in the model for WAI. All other variables were excluded for the final models.

4.1Strenghts and limitations

The strengths of our study are that we followed regular occupational health services of a large, heterogenuous working population (n = 428) and a variety of companies, increasing the likelihood of generalizable results. Additionally, we studied two measures of resilience.

Observational research also has weaknesses that can affect results. One limitation of this design was a potential for selection bias. Nine companies chose to follow-up within 24 months and were therefore excluded from the study. At baseline, WAI and UBES-9-vitality scores between companies excluded for this reason, did not differ significantly from the included companies. Therefore we concluded that unfavourable score on WAI or UBES-9-vitality at baseline did not influence the follow-up period chosen by the companies. Another potential selection bias was set by the option to use a different set of questionnaires for corporate motives, which occurred in four of the remaining companies. To examine whether there was selection bias by not administering these questionnaires at follow-up to workers with an undesirable WAI or UBES-9-score, we tested for a difference at baseline for workers with and without WAI or UBES-9-vitality variables at follow-up. There appeared to be no statistically significant differences, and this loss to follow-up was unlikely to be related to baseline WAI or UBES-9-vitality score. It is likely that sampling bias has influenced results because workers on sickleave were not included, causing a healthy-worker effect. Previous studies have demonstrated that poor health [37], sickness absence and presenteeism [38] were associated with unsustainable employability. The excluded workers in our study were lower educated and had more physically demanding work, which are known risk factors for reduced sustainable employability [37, 39]. Missing workers at high risk of unsustainable employability, might have decreased the explained variance we found due to the ceiling effect of the WAI and UBES-9-vitality-score. Finding a modest, but significant explained variance in a selected group of healthy workers, can be seen as a promising finding for early detection of the presence of a protective factor for sustainable employability. Methodological bias has likely influenced the positive relation we observed between vitality and the BRS-DLV.

As mentoined before, mental resilience is part of the definition of vitality and it is highly likely that it associates with resilience measured with the BRS-DLV. A conceptual difference between the two is that UWES-9-vitality explicitely focusses on vitality at work and the three questions do not seem to reflect the definition of resilience, while the BRS-DLV is developped to measure resilience in general [9]. It is possible that workers are vital for work in early stages of health conditions. In the recent dynamic view of health as ‘the ability to adapt and to self-manage, in the face of social, physical and emotional challenges’ [40], the human capacity for resilience and coping with new situations is highlighted [40]. Therefore we decided to screen for general resilience as measured with the BRS-DLV for early prediction of workability and vitality. A final confounder could be that the influence of lifestyle activities to HRV results (e.g., poor sleep quality, alcohol consumption or coffee intake [27]) cannot be ruled out, even though participants were asked to avoid them prior to testing.

The field of studying factors contributing to staying at work (which is not the opposite of risks for leaving work) is relatively young, but very important for wellbeing of the aging workforce [2, 31]. With more knowledge about stay-at-work factors, early interventions could be developed or applied to support workers. This is parallel to managing risk factors for leaving work, which is already performed in most companies. Future studies with longer follow-up periods in large cohorts are required to gain the necessary insight into factors supporting workers. Studies with longer follow-up periods might also be of additional value to detect the sensitive periods when interventions are most successful.