The intervention protocol was reported in full elsewhere (see 16).
Setting
The study was conducted in Pihlajalinna Työterveys, a large private occupational health services’ provider, which at the time of starting the study had 28 private healthcare units across Finland. In 2015, Pihlajalinna Työterveys had approximately 68370 employees on their register. The organization went through several rounds of mergers and corporate acquisitions during the study period, which led to a substantial increase in patient and health care unit numbers.
The intervention
The intervention was multifaceted and implemented sequentially. First, a notice was sent to the entire organization informing all practitioners that the study would be conducted. The intervention consisted of two separate activities: one, training, mentoring and follow-up of physicians in intervention units on how to identify and record work related illnesses during primary care visits and how to identify and record risk of work disability. The training sessions were conducted at each intervention unit. During the sessions the intervention and its components were introduced, and information about work related illnesses was reinforced. This also included training on the intervention processes - actions that were to be initiated after a patient was identified during a visit as at risk of work disability. Following the training, a project physician responded to questions and followed up with training participants telephonically. Second, an organization-wide change was made to the electronic health care system, clarifying the way in which work relatedness and risk of disability pension was recorded. This change reinforced the messages given in the training. The electronic change was made to clarify language in sections where work disability risk was assessed. No specific training was conducted on the change to the electronic record system, as changes were minor and had been introduced in the intervention training. Therefore, intervention physicians were more likely to adhere to the change in the electronic health record as they had been trained.
The trainees in the intervention sites were those occupational health physicians, who were responsible for collaboration with their own client companies, working at any of the 22 sites included in the study. The intervention and control sites had 58 physicians and 50 physicians, respectively, employed during the study. These physicians would be responsible for contacting workplaces, and to be involved in tailoring clients’ work tasks or conducting other workplace interventions.
If the OH physician noted a client’s visit as being related to work or that the client presented with a situation that could potentially result in work disability in the near future, they marked this onto an electronic system. Following this, a sequence of events was kicked off at the intervention sites. The OHS nurses responsible for the employer organisations collected the patients identified as at risk from the electronic health record. They then initiated the interventions that a physician recommended together with physicians, either for the clients or more widely at the workplace. These interventions could include e.g. an occupational health collaborative negotiation to modify the employee’s work tasks or timing, or organizational interventions focused on workplace ergonomics or teamwork counselling. Other interventions could include e.g. starting medical or vocational rehabilitation for the individual patient, which involves both the workplace and the client/employee. It was not possible to collect the number of these interventions in this study since the individual patients that were identified as at risk of work disability or that had a work related condition could not be associated with the interventions conducted at workplace level. A fuller description of the intervention can be found in the TIDIER reporting guide for population health interventions: tidierguide.org/#/gen/pFqrFqw3M
Information about the study was sent to all sites in April 2016. The intervention training was conducted in May 2016. The electronic change to systems was implemented in 9.3.2017. Data collection ended in December 2017.
Randomisation
We included 22 Pihlajalinna Työterveys clinics that were functional in 2016 in the study. We treated each healthcare unit as a cluster, as individual randomisation in this context would have been challenging. NT, the team statistician, conducted initial simple randomization to randomize the first four clusters. After this, we used the minimisation approach to randomise the remaining 18 clusters so that the following confounders balanced across the intervention and control sites (see 16): 1) the occupational sector (e.g. industrial, service sector, public service), 2) presence of a large industry client, and 3) client volume per site. The occupational health professionals and the research team were not blinded to the intervention.
Outcomes
Our primary outcome was reduction of the mean number of medium term (4-14 calendar days) of sickness absences per intervention and control centre from baseline after 1 year of follow up as measured by records kept by OHS patient records 16. We considered medium length sickness absences from 4 to 14 calendar days, instead of 9 working days as indicated in the original protocol. We chose this to match our findings more closely with the Finnish Insurance Agency’s definitions for sickness absences, which considers medium length sickness absences as absences including 9 working days (including Saturdays but not Sundays). The patient records included also weekends as sickness absence days, which differed from this approach. A similar choice was made in 17.
Our secondary outcomes were
- Reduction in mean number of short term (1–3 consecutive calendar days) sickness absences from the workplace per cluster after 1 year from the start of the intervention as measured either by self-reported sickness absences recorded on the OHS system or sickness absences certified by OHS physicians working at the included OHS units.
- Reduction of mean number of any form of work disability pensions as measured by an employee registered as receiving a work disability pension on the central pensions register from baseline to up to 2 years from the intervention
- Reduction of mean number of long-term (15 or more consecutive calendar days) sickness absences from the workplace per cluster from baseline to 1 year after the intervention as measured by OHS records
The follow-up time was set at 1 year due to funding and planned study duration. This article focuses on reporting the primary outcome, medium term sickness absences, as we deemed the follow-up period too short to report on disability pensions or long-term sickness absences. We also report short-term sickness absences and the process indicators collected on recording consultations’ work relatedness and risk of work disability across control and intervention sites.
Power calculation
Our initial power calculation suggested that we would have 91% power to detect a 10% change in mean sickness absence rates across intervention and control clusters, if we had 22 occupational health units with 24892 patients. For the trial, we retained all 22 units, with 26804 patients recorded on the system.
Data collection
We collected medical record data on patients’ healthcare consultations at Pihlajalinna Työterveys from 2015 to 2017. The medical records included 68370 patients in 2015 and 107413 patients in 2017. The cohort was dynamic, in that patients could be added to the cohort as the study progressed. Data were pseudonymised, and researchers had no access to patient identifying data. All patients above the age of 18 and whose employers had a contract with Pihlajalinna Työterveys including primary healthcare services were included in the study.
The data were combined with pseudonymised data from the Finnish Centre for Pensions, where we obtained all participants’ pensions granted for the study period.
Data analysis
After data collection was complete, we noted that Pihlajalinna’s acquisition of another large occupational health services provider impacted on our outcomes. Therefore, we used all initially randomised sites in the intention to treat analyses and excluded them in the per protocol analyses.
We included data on curative patient visits to OHS physicians responsible for client organisations. OHS services have many casual workers, who deal with primary care patients but are not occupational health specialists and most of them were not exposed to training. We also excluded preventive visits such as health examinations. We analysed data 6 months before the intervention, during the intervention, and after the intervention for 6 months. After initial analysis we chose a period of 6 months after the intervention corresponding with the same season of the 6 months preceding the intervention, to ensure that seasonal effects did not confound our analysis. We analysed data using ANCOVA, setting alpha at 0.05.
We also analysed process indicators among intervention and control clinics. These indicators included whether physicians had changed their practice of recording a consultation’s relatedness to work after the intervention. The intervention required a physician to record whether or not the patient’s visit was related to work or whether this was not assessed. We analysed changes after the educational intervention and after the change in the electronic system using descriptive statistics.