The present investigation observes a cohort of trainees, from the start of training till the first year of employment. The proportion of limited HL (problematic or inadequate HL) was 47%. As regards the health behaviour, there were high prevalence values for unfavourable nutrition (51%), smoking (26%), low physical exercise (67%) and risky alcohol use (30%). As regards health indicators, 42% of persons were overweight, 17% reported poor subjective state of health and 42% of persons exhibited low psychological well-being. No longitudinal associations were found between HL and health behaviour. A statistical significant cross-sectional odds ratio of 3.2 was observed for the association with psychological well-being.

For HL, there was a slight increase in the mean value over time. This was particularly marked in the health and welfare services sector group, but was not statistically significant. The level of HL at T0 was higher for employees in health and welfare services than for persons from the sector office/sales/technology – presumably due to selection. The reason is evidently that the various courses for training in the health and welfare services gradually lead to greater increases than in the comparative group; these increases are statistically significant at time point T3. There were nevertheless heterogeneous changes in HL over time. Thus, a decrease in the HL of at least one point was observed in 38% of participants. This may indicate a bias due to a faulty subjective assessment at T0.

The observed prevalence of limited (problematic or inadequate) HL was 47%, which is less than the values of 53% and 49%, respectively, found in the baseline and follow-up T1 publication and is lower than data from the German speaking area [24, 35]. In the recently performed survey in the adult German population, this value was 59% [10]; in 2015, Jordan & Hoebel observed a prevalence of 44% [36]. For younger subjects, the proportion of limited HL has been reported to be 58% (15-year old Austrians) or 69% (students in a German vocational college for health) [25, 37].

A statistically significant reduction in risky alcohol use was observed in both groups of sectors and represents a continuation of the trend observed in the T1 publication of the cohort [35]. Nevertheless, the prevalence value of 30% was clearly above the comparative German data for the age group of 18 to 29 years, where the gender-specific values were 14% for men and 18% for women [38]. The increase in BMI over the time in this age group is consistent with the data from German health reports, even though the value of 25.3 found in this study was above the reference values in the official statistics (20–24 years: 23.5, 25–30 years: 24.5) (39). A possible explanation for this increase could be the restrictions on activities in the two lockdown phases in Germany during the observation period of the study (1st lockdown: March 2020 to May 2020, 2nd lockdown: December 2020 to May 2021) However, due to the lack of a comparison cohort, this explanation is purely hypothetical. The prevalence of poor subjective state of health was 17%, which is clearly greater than the comparable values in the GEDA 2014/2015 EHIS survey, where the gender-specific prevalence were between 2.5% (women) and 1% (men) in the age group of 18–29 years [40]. In addition, a large section of the subjects (42%) reported poor psychological well-being at T3. It is certainly possible that the generally high values for unfavourable health behaviour (unfavourable nutrition: 51%, smoking: 26%, low physical exercise: 67%, risky alcohol use: 30%) and poor psychological and physical health are at least to some extent consequences of the stress from the covid-19 pandemic in Germany, with the restrictions to social contact in 2021, and should be regarded as a period effects of the pandemic.

There were striking changes from T0 to T3 in the workplace-related variables of recognition, satisfaction at work and commitment. These changes depended on the sector, namely that in the health and welfare services sector there were statistically significant deteriorations over time in all three parameters. On the one hand, the mean for recognition tended to increase during the training period (T0-T2) for health and welfare services, with a slight decrease in commitment (neither statistically significant). On the other hand, for the period T0-T3, the values for recognition, commitment and satisfaction at work deteriorated in this group of sectors. This indicates that for the group health and welfare services, the transition from the training period into actual employment is more difficult with respect to the workplace parameters examined here. It may also indicate that the period effect from the covid-19 pandemic is more marked for employees in social sectors than in other sectors. Daily care of patients and clients in the context of a pandemic is more stressful, due to the inevitable contact restrictions, distancing and the relatively high risk of infection. This may all cause entrants to have doubts about their choice of profession. It is difficult to assess the specific mean values we found for the COPSOQ scales, as the groups include different subsectors.

No associations were observed between HL at T0/T1/T2 and the various indicators of health behaviour at T3. An alternative approach was to consider that HL might act immediately on health behaviour, so that HL at T3 should be compared with health behaviour at T3 – but here too no associations were found. This negative result is consistent with the lack of association between HL and health behaviour in the T1 publication [35]. On the other hand, this lack of association between HL and health behaviour is, to some extent, incompatible with the results of previous studies. In a systematic review, Fleary et al., reported that 13 of 17 studies on adolescents found statistically significant associations between HL and health behaviour [17]. A German study on adolescents with limited HL employed the long version of our questionnaire (HLS-EU-Q47) and found no associations with tobacco and alcohol use, but associations with nutrition and exercise. Moreover, a study with 15-year old Austrian adolescents failed to find an association between two out of three indicators of health behaviour (alcohol use and smoking; the association with exercise was weak (r = 0.14)) [25]. On the other hand, a systematic review of interventions in adults found a clear association between HL and the outcome health behaviour [41]. Thus, published reports on this association are inconsistent. In the present study; the lack of association between HL and health behaviour may be because our subjects wrongly assessed their HL, due to their relative youth. This would indeed be consistent with the decreases in HL over time in our study. Another possible explanation would be that the subjects accept the risk of unhealthy behaviour, even though they know better. In focus groups, Joseph et al. considered that HL provided an instrument for making well informed decisions. Nevertheless, the participants admitted that they sometimes failed to exploit these insights by adopting risky behaviour [42]. Another possible explanation is as follows. In a mediation analysis, one study with the Chinese version of the 47 item HLS-EU-Q47 concluded that only the subscale Use Information had an indirect effect on health through health behaviour [43]. It seems plausible that the content of this subscale may be more directly related to health behaviour than are the subscales Find Information, Understand Information or Assess Information. We were unable to replicate this analysis in our study, as we used an item-reduced version of the HLS-EU-Q16.

For psychological well-being, an increased cross-sectional odds ratio of 3.2 was found for inadequate HL and of 1.8 for problematical HL (not statistically significant). This confirms the trend in the results in the T1 publication [35]. This result was also confirmed by Björnsen et al. in a study on adolescents [44]. In a cross-sectional study with Norwegian (male and female) adolescents aged 15 to 21 years, it was found that high mental HL was associated with high psychological well-being. This was confirmed in additional cross-sectional studies with adults. Zhang et al. observed this association in a population-related study in Hong Kong, analogous results were reported by Fiedler et al. in German industry managers and by Amoah et al. in a population-related sample in Ghana [16, 45, 46].

There were cross-sectional associations between HL and subjective state of health, although the estimates were not statistically significant. Employees with inadequate HL have a 2.7-fold greater probability of lower subjective state of health than those with adequate HL; there is a 2.2-fold difference for persons with problematical HL. This result confirms the statistically significant results in the T1 publication [35]. Little has been published on the effect of HL on subjective health in this target group. Two studies with vocational students show that young adults with high HL also have better subjective health than students with low HL [47, 48]. In a systematic review, Sansom-Daly et al., concluded that there had been very few studies in adolescents on the association between HL and various health indicators and that the results of these studies were inconsistent [19]. On the other hand, a large population-related study in China demonstrated unambiguous associations between HL and subjective state of health [49].

The associations were only found in cross-section and this indicates that there is a direct temporal relationship between HL and health in the data. The lack of longitudinal associations may be due to the heterogeneous time courses and indeed the frequent occurrence of reductions in HL may be regarded as an indication of subjective misassessment at the time point T0. Adolescents may have had little experience of illness or of the health system due to their age and may exaggerate their own knowledge. Some simply do not have the expertise to assess their own health expertise. Nevertheless, longitudinal effects were observed in the T1 publication, which contradict the above argument. However, the fact that these associations were also observed in the T1 cross-section indicates that HL correlates better with itself over short intervals than over long intervals (T0-T3) (Unpublished results). In summary, the null hypotheses must be accepted here due to the observed results in the longitudinal analyses. We cannot say to what extent this result can be explained by an excessively small sample. But there is an indication in the data that the instrument for assessing HL is not optimal for the underlying age group. In future studies, it is important to consider to what extent the assessment of objective HL is more meaningful in research on adolescents than the use of an instrument that measures subjective HL, such as the HLS EU 16. In his work, In his work, Okan shows the possibility of using both subjective and objectively recorded HL in children and adolescents [50].

For a group of entrants in the first year of employment during the covid-19 pandemic, the present study shows that preventive measures are needed for high risk behaviours, overweight and poor psychological well-being. The results also show that HL is a modifiable parameter and that it is associated with health in young employees. In particular, trainees who are not in the health services or social welfare should receive more instruction on health during their training. It would perhaps be best for these sectors if the vocational colleges could assume the responsibility for this task, as training centres that are remote from patients and clients tend not to emphasise themes related to health support and prevention – apart from health and safety protection. Thus, for example, the training curricula could contain information relevant to health and emphasise personal prevention – all in addition to health and safety protection. The objective over time is then to enhance the trainees‘ HL.

The relative low response rate and the relatively high proportion of drop-outs increase the size of a possible selection bias in the present study. Although the design of the baseline survey included a full survey of the target group in the study region, the study sample is not representative at T3. In the drop-out analysis it has been observed, that participants with an unfavourable health behaviour had a double risk for a drop-out. So the prevalence of unfavourable health behaviour may be underestimated in the follow-up cohort. As the proportion of men at T3 is low, it is not meaningful to perform a gender-specific evaluation. Thus, it is also impossible to draw gender-specific conclusions about preventive measures. Because of the low number of cases, caution is needed when interpreting the observed prevalence. Because of the number of cases, individual sectors were combined, even when the conditions at work are different. This is particularly the base for the group office/sales/technology. In the hypothesis testing for the association between HL and health indicators, only cross-sectional indicators could be found. Thus, these findings have a lower level of evidence than associations between longitudinal data. Cohort analysis over all the time points of the survey (T0, T1, T2, T3) was not considered, due to the small number of cases, so we decided to present data for the longest follow-up period between T0 and T3 including the transition from education to the first year of work. In the longitudinal analysis there were no significant associations observed, but we cannot rule out that this was due to the small sample size. Moreover, the independent and dependent variables are from the same source, so that a bias due to common-method variance cannot be excluded. All information is based on self-reported data obtained from a questionnaire. It should also be remembered that the results were influenced by the simultaneous occurrence of the COVID-19 pandemic during the training period. This is of special relevance for employees in the health services and in health and welfare. It is possible that the observed work-related stress and demands would have been lower than during a pandemic-free period, Nevertheless, this study contains historical data, that reflects professional stress suffered by trainees during their transition to the first year of employment during the pandemic.