We conducted a systematic review and meta-analysis to examine the risk of SARS-CoV-2 infection in various nonhealthcare occupations. A total of 25 publications were included, mainly describing the first year of the COVID-19 pandemic. Meta-analyses were performed for 20 occupations other than HCWs, predominantly showing an increased risk of exposure to SARS-CoV-2.

For seven occupations we identified a statistically significantly increased risk of being infected with SARS-CoV-2, which we classified as moderate to high evidence. All the occupations identified in our study as belonging to the ‘food and hospitality’ sector (2923; 2930; 6330) and to the ‘social services’ sector (8312) constitute this category, complemented by individual occupations in the ‘transport and logistics’ sector (5211; 5133) and the ‘security and surveillance’ sector (5311). Two further occupations, attributable to ‘security and surveillance’ again (5318) as well as to ‘cleaning’ (5411), showed a risk for infection significantly increased, however, they were rated as low evidence. These occupations cover three of four sectors specified as employing ‘essential workers’, which were defined by Castellazzi [42] as ‘agriculture’, ‘health and personal care’, ‘transportation’ and ‘food distribution’. In Germany, ‘law enforcement agencies’ are considered essential in addition to ‘food’, ‘transportation’ and ‘emergency and rescue services’ [4]. The majority requires personal contact in their execution.

For another ten occupations, we identified a risk not statistically significant and rated with low evidence, predominantly showing an elevated risk. Though the vast majority can be considered as ‘essential workers’ as well, the application of protective measures was possible: In some occupations, such as ‘bus and tram drivers’ (5213) as well as ‘sales occupations (retail) selling foodstuffs’ (6230) and ‘cashiers and ticket agents’ (6211), the potential for high infection rates was anticipated due to high numbers of customers. As preventive measure acrylic glass shields were installed as a technical barrier. In certain time frames, the reduction of maximum number of customers in grocery shops further enabled the mandatory regulation of distancing while in public transport the overall number of passengers was reduced due to lockdown. A high risk was also expected in occupations in contact with children, whereupon several countries included ‘occupations of child care and child rearing’ (8311) as well as ‘teachers in primary education’ (8411) in the organisational measures for closure (cf. [5]). However, the closure of nurseries and schools might have brought their occupational risk into line with that of other homeworking occupations.

For a single occupation, we identified a statistically significantly decreased risk of being infected with SARS-CoV-2, rating it with moderate evidence: Even though ‘accounting’ can be attributed to ‘financial services’, which is listed as an ‘essential occupation’ [4], it can be performed digitally, avoiding any personal contact.

As only observational studies were included in the systematic review and subsequent meta-analyses, the results represent risks that were influenced by both individual behaviour and policy-related factors. Thereby, our results do not represent a raw risk, but should be read in the context of policy measures applied. In addition, different periods of policy measures in the countries contributing to meta-analyses may further explain ambiguous findings.

Occupations described in our study predominantly account for the tertiary sector, with ‘food-processing’ being the only occupational group representing the secondary sector; occupations of the primary sector do not appear. Though ‘agriculture’ [42] or ‘farming, fishing and forestry’ [5] are termed consistently as ‘essential’ or ‘frontline’ occupations, we did not find enough datasets in the eligible literature to perform a meta-analysis.

Meta-analyses were calculated predominantly for occupations of lower social status. A reason for this focus can be found in aspects of data protection, as very small numbers in cases led to non-reporting in the original studies [40] and applies in occupations with low numbers of representatives, including leadership positions. This is in contrast to results based on the German National Cohort [43], where a remarkable number of occupational sectors of higher social status were reported as having a higher infection risk in the first wave.

The heterogeneity perceived as ‘considerable’ (I2: 75-100%) to ‘substantial’ (I2: 50-90%) might be explained by the consideration of different outcomes, the dynamic rate of new infection and the different implementation of prevention measures in the countries under consideration.

Four out of eight domains of GRADE did not result in a modification of the rating. As all studies considered working adults, ‘indirectness’ did not need to be addressed. Due to the pandemic event in which every article on COVID-19 was published, predominantly with ‘open access’ [44, 45], no ‘publication bias’ was observed. The binary outcome, i.e., virus detected or not, detained from assessing a possible ‘dose-response’ relationship. Finally, ‘under-estimation’ was not observed.

The definition specified a priori for downgrading in the domain ‘risk of bias’ seems too strict to us in retrospect. As it refers to a difference between results of low RoB and high RoB, it does not take into account whether results of low RoB tend towards a strong or weak relationship. Thus, we might have downgraded a result, despite its subgroup result of low RoB exceeded the overall result.

The methodological quality of the included studies varied considerably.

In terms of ‘exposure’, the accuracy of the occupational description ranged from self-declaration to information provided by the employer according to a classification system. The former was sometimes easier to translate into a standardised form than the latter, because translation tools offered more than one suggestion. Thereby, the differentiation appeared to be very simplistic, because different activities – and therefore exposures – could not be taken into account. Although we have taken into account that information on current employment may be out of date if the data collection was more than 12 months before infection, there may have been other misclassifications, e.g., inaccuracies in death certificates.

In terms of ‘outcome’, we considered three stages of rising severity. While for the infection itself the availability of testing capacity was a crucial factor leading to a potential bias in the results [40], mortality is influenced by many additional factors such as pre-existing illnesses.

Though ‘age’ and ‘sex’ were known as confounders in an early stage of pandemia [46], most of our eligible literature did not take them into account. This lack, however, is not only a disadvantage in observational studies but was detected in clinical studies as well [47]. Comparably, information on socio-economic status (SES) was not recorded in all our included studies, for which reason a confounding effect on our results cannot be precluded. Furthermore, studies published within our search period did not address comorbidities. A recently published case-cohort study using German statutory health insurance data did include comorbidities in its analyses [48]. However, their inclusion did not explicitly change the conclusion but generally supported our findings.

The strength of sex as an influencing factor depends on the stage of the outcome. While infection itself showed no sex differences, it was found pronounced in case fatality rate [46]. As its marked focus was identified in the age groups 50–59 [46] and 60–69 [49], this is related to the working population. Thus, results derived by mortality data will probably overestimate the occupational risk in professions predominantly executed by elderly male when no adjustments for sex were applied (cf. [40]).

In times of pandemia, it is hard to find an unexposed group considerable as comparator. Hence, we used a hierarchically structured approach of ‘little exposed’ or ‘less exposed than’. Of the four possible categories defined in the PROSPERO protocol, three were covered in the extracted literature. In the majority of the literature, a non-exposed comparison group was used (e.g. [50]). However, in some cases, the group assumed as most exposed (HCWs) was regarded as comparator (e.g. [51]). Taking an exposed group as a comparator, results in underestimating a risk.

Furthermore, the subgroup analysis regarding ‘risk of bias’ revealed a ‘bias towards the null’ for studies rated as high RoB. Again, this results in underestimating a risk.

Of the 25 studies included, 20 were classified as studies with a high RoB. This was mainly due to weaknesses in the evaluation of the data, such as insufficient consideration of confounding factors and inappropriate analytical methods. In addition, there were weaknesses in the recruitment of participants and the definition of exposure. Only five of the included studies showed a low RoB, while at least one minor bias factor was present, as well. As a rare situation, we had to exclude one of these studies [29], as it overlapped substantially in area and period with another study rated as low RoB [40].

We were able to extract 76 occupations from the literature at KldB-level 4. However, only for a restricted number of 20 we can report results, while two-thirds of the occupations were available in only one or two data sets. This can be attributed to the underlying studies’ own methodological aspects as some focussed on ‘essential occupations’ explicitly [18] or took into account the expected level of exposure [38], both resulting in a focus on service occupations. Other studies followed statistical considerations and excluded occupations with low numbers of employees or low numbers of infections [26, 35, 40], which leads to a limitation to occupations with higher risk of infection. Outbreak studies do not contribute to our result.

Though some studies defined their occupational exposure according to classification systems [18, 35, 37, 40], the majority did not. Thereby, we were not able to transfer all our data to the international acknowledged classification of the International Classification of Occupations (ISCO). The latter follows a skill level with level of leadership as first grouping, which was not indicated in the extracted data sets. In contrast, the German classification starts with the sector and narrows it down to educational aspects on level 5 (which we did not consider) [10, 52]. This enabled us to assign each data set according to its potential depth.

While a detailed occupational description was desirable, it has the disadvantage of scaling down the number of cases. Furthermore, similar occupational tasks – such as passenger transportation in busses and in taxis – are separated though they might pose a comparable risk.

Finding more than 9.000 texts within a search frame of two years are a massive output for a topic as young as COVID-19, even more as occupational risk is often a side issue. We refrained from performing forwards or backwards citation tracking [53] due to a very limited time interval (i.e., 01.01.2020–01.02.2022). Instead, to include all upcoming publications, we searched medRχiv for relevant texts, a database that rapidly gained attention during the pandemic [54]. As all texts retrieved from this source were peer reviewed by the time of data extraction, the confidence in their quality was given [55].

The majority of the texts excluded in title/abstract screening focussed on HCWs. This emphasis has been assessed in the literature before [40]. Not only their risk of infection was a subject matter but also their status of mental health [56,57,58,59].

For all texts eligible, the pdf was retrieved without problem for full-text screening. This probably is due to an understanding of cooperation [44, 45], issued by the scientific community at the beginning of the pandemic, to make research results available free of charge.

Though we conducted our search at the end of the second year of the pandemic, the reported data of the included studies pool in the first year. As vaccination only occurred in the end of 2020 and by then was offered exclusively to selected groups not being of interest in this study (i.e., HCWs, elders, and vulnerable persons), we covered a period in which the effect of vaccination is neglectable.

While studies giving insights on pandemic impact in African countries were rare but existed [60,61,62], there were none from Oceania. This probably can be explained by their isolated position and their rigorous border controls, especially within the first year. Thereby, spread within occupations was scarce and not worth doing research on.

We examined the risk of SARS-SoV-2 infection in occupations outside of the healthcare sector systematically. Search for literature was conducted worldwide and in all languages. Studies not available in German or English were translated professionally for full text screening (i.e., Spanish, Italian, Persian). Thereby, we are able to present a full picture of the scientific literature referring to the first year of COVID-19 pandemia, enabling us to describe the risk of infection without the effect of vaccination.

We describe the risk of infection for a broad variety of occupations contextualising service occupations. They cover a huge number of workers and were assumed to have a high risk of infection because of their professional activity.

We are not able to give information on occupations in which few people were employed, such as rare professions or leading positions. Furthermore, self-employed occupations are underrepresented.

Within an occupation, different tasks and activities are performed, of which some pose a higher risk for infection than others. The varying risk of infection rather depends on whether these tasks are performed or not than on being employed in a certain job. However, as information as detailed as that was not assessed in times of the pandemic, occupation is a suitable proxy.

In analysing the first year of pandemia, we integrated at least two waves. Studies considering them separately showed marked differences in occupational risk [63, 64].

In COVID-19 pandemia, the workplace became a field of public health. Protective measures implemented by health policy ranged from personal protective equipment to extensive lockdown, and had an impact on workplace safety and infection rates. It differed between industries and countries, and essential workers were protected from SARS-CoV-2 infection to a varying degree.

As home-office has gained popularity since then, health policy measures of the pandemic have long-term effects in the workplace. The experiences made in the pandemic should be further analysed in order to be able to make recommendations for action in the future.