Validity of self-reports on knee loading
The present study showed two different aspects of self-reported knee load: good to acceptable quality in identifying knee postures but mostly poor to very poor quality in quantifying the load. These conclusions are supported by related studies on several musculoskeletal risk factors (Descatha et al.
2009; Stock et al.
2005; Unge et al.
2005) and knee loading in particular (characteristics of the referred studies are shown in Appendix C in Supplementary Material): In a Finnish study on forest industry workers, Viikari-Juntura et al. (
1996) described a poor correlation between observed and self-reported amount of kneeling and squatting (Spearman’s ρ = 0.42,
p < 0.001). Hence, they determined self-reports to be helpful in identifying high exposure groups but to be inappropriate in quantifying the exposure. Their results were based on the direct workplace observations of 36 workers, compared with self-reports on the exposure of an average work shift from 2,756 workers.
Baty et al. (
1986) examined working postures of 46 nurses by observation and registration of major body postures every 15 s. At the end of the work shift, participants were asked to assess the amount of time spent in several postures. For kneeling and squatting, a good agreement between observed and self-reported occurrence was found (22/23 and 10/11 agreements, respectively), while the nurses overestimated their duration of kneeling and squatting four times on average. It should be kept in mind that kneeling and squatting postures occurred only infrequently.
In a Dutch study, 35 mechanical repairmen were observed at the workplace and asked to keep a log every hour to assess exposure to several musculoskeletal risk factors (e.g. kneeling/squatting) for a whole work shift (Burdorf and Laan
1991). Subjects were able to assess the occurrence of kneeling/squatting activities quite well, but the percentage of daily work time in these postures was slightly underreported.
In a German study, task analyses on 25 workers were carried out using an observational method (Klußmann et al.
2010). At the end of the work shift, 92 % of the subjects were able to report the occurrence of knee postures correctly but failed in quantifying their kneeling exposure (average deviation between self-reported and observed duration of kneeling, 171 %).
In another German study, 75 construction workers were observed for 4 h at the workplace, and their exposure to kneeling and squatting was quantified with a stop watch (Bolm-Audorff et al.
2007). After the observation, subjects were asked to assess the duration of kneeling and squatting postures during the observation. The results of the self-reports and the observation showed a good Pearson’s correlation (
r² = 0.74,
p < 0.01), but workers seemed to overestimate their knee load systematically: the median self-reported duration of knee postures was reported as 35 % of the working shift, while the median for the observations was 21.9 % (
p < 0.001).
However, there are a few studies on this topic with contradictory results. In a British study with 123 participants from various occupations, the self-reported durations of kneeling postures taken directly after the examination agreed well with the observed amount of kneeling (Pope et al.
1998). This may be caused by the relative rare occurrence of kneeling activities (only about 50 % of the observed tasks included this exposure) and the observation method (recording of postures all 30 s during 1 h of working time), which may not be suited for quantitative measures of highly dynamic tasks. A Danish study on occupational knee loading in 33 floor layers and 38 carpenters also reported good correlations (Spearman’s ρ = 0.89) between self-reported and video-recorded amount of kneeling and squatting (Jensen et al.
2000). However, the examined working sequences were rather short (three to 30 min) and included very homogenous tasks, which may support a good recall of the knee load.
The variability of the studied exposure to knee-straining postures may also have an impact on the validity of assessment. In comparison with the referred studies above, our study sample must be seen as rather homogeneous in respect to knee-straining postures (CV = 0.72, cf. Appendix C in Supplementary Material) as we involved tasks in our study which were supposed to be knee-straining.
All reported studies examined only self-reports taken immediately after the exposure event or at the end of the working shift. In contrast, the present study was interested in subjects’ ability to assess their exposure a half-year later, as well. In this second survey, subjects’ ability to recall the occurrence of knee postures can be rated as acceptable to good. However, the validity of the self-reported durations of these postures was worse than in the first survey. To the best of our knowledge, there are no similar published studies on this topic.
Assessment behaviour and impact of exposure level
In both surveys, participants tended to overestimate their exposure, especially in survey
t
1 (87.2 % overestimations). Nevertheless, underestimations can be observed in both surveys. Both phenomena have been reported in several studies on assessment of knee loading: clear overestimation (Bolm-Audorff et al.
2007), predominating underestimation (Burdorf and Laan
1991), and deviations in both directions in one sample (Jensen et al.
2000). Thus, the assessment behaviour may depend on the wording of the questionnaire, the study sample, or the exposure level (Barrero et al.
2009).
As this study indicates, exposure level seems to have an enormous impact on the validity of self-reported knee exposure. In both surveys, differences between reported and recorded durations of knee postures were small at a low exposure level but increased with increasing exposure. Participants were able both to report the absence of knee postures exactly and to assess short time exposure, especially by comparing absolute values (see Bland–Altman plots) rather than relative ones. On the other hand, high-exposed subjects were misjudging their amount of knee loading by far. Confirming this effect, a study on the duration of computer use of 87 computer workers reports comparable assessment behaviour for low- and high-exposed subjects (Heinrich et al.
2004). But in contrast, another study on that topic gives an opposite result: agreement between self-reported and observed duration of computer use of 572 office workers improved with increasing exposure (IJmker et al.
2008). This effect might be explained by the use of categorical data (seven response categories for hours of computer use per day), while we used continuous data for assessment in our study. With respect to occupational knee load, only one of the cited studies took assessment behaviour of low- and high-exposed subjects into consideration (Klußmann et al.
2010). In a sub-analysis of this study, high-exposed workers showed a better ability to assess their exposure than low-exposed. However, study sample was rather small (
n = 25) and deviations between both methods were only reported as relative differences instead of absolute numbers; thus, the effect may be overestimated.
Impact of knee disorders on the validity of self-reports
The present study gave no hint of a differential misclassification of exposure due to self-reported knee complaints. Participants both with and without such complaints showed comparable assessment behaviour. This result seems to be contrary to studies reporting differential misclassifications caused by several forms of musculoskeletal complaints and risk factors such as low back pain and manual material handling (Wiktorin et al.
1993), neck-shoulder complaints and awkward postures of head, back and arms (Hansson et al.
2001), or upper limb complaints, and physical activity (Balogh et al.
2004).
In terms of occupational kneeling or squatting, only a few studies considered the impact of musculoskeletal disorders on the assessment behaviour. Moreover, if complaints were taken into account, it was not about knee complaints. Burdorf and Laan (
1991) found no impact of low back pain or shoulder pain on self-reported kneeling or squatting of mechanical repairmen. Sample size of that study was certainly small (
n = 35) and kneeling or squatting just made an average of only 14 % (SD, 12) of the observed time in the sample. In contrast to that, Viikari-Juntura et al. (
1996) reported an increased risk of reporting high workload for forest industry workers having severe low back pain, e.g. for kneeling and squatting (OR, 1.6; 95 % CI, 1.2–1.9). Again, sample size was small (18 subjects with and 18 subjects without low back pain), and squatting or kneeling was rare in both groups (median, 0.0 h each). As the present study has dealt with knee complaints, our results cannot be closely compared to those studies. Moreover, our study concentrated on kneeling or squatting tasks (median, 32.7 min or 29.7 % (0.0–92.7) of knee postures per measurement). With certain constraints, it should be noted that subjects with severe knee pain probably did not participate in our study due to sick leave.
Study limitations
The present study has several limitations that should be considered when interpreting the results.
The study was based on the voluntariness of participation of companies and subjects, which might have led to selection bias. Moreover, we examined only tasks where we expected knee-straining postures. Thus, our results are not representative for the whole working content of the examined trades.
While in survey
t
0 all measured subjects filled out the questionnaire, in survey
t
1, only 65.8 % of the participants responded. However, compared to response-rates of other studies in Germany, this can be seen as quite successful (Latza et al.
2004). A non-responder analysis yielded similar to identical characteristics for responders and non-responders (see Appendix B in Supplementary Material). This lack of difference suggests that the lost to follow-up may not be an important issue, and the risk of a non-responder bias may be ruled out.
As the second survey was conducted by mail, study participants were only able to ask comprehension questions in the first survey when study staff was on site. Thus, comprehension problems may have occurred in the second survey more often and may have biased the exposure assessment, for example by self-reported exposure wrongly related to a whole work shift, rather than to the measuring period. However, we attempted to minimise this effect by using the same questionnaire as in the first survey, accompanied by information on how to correctly fill it out. In addition, we gave a short description of the work performed during the exposure measurement at t
0. This procedure could have artificially reduced recall bias as such information cannot be provided in an epidemiological study, for example.
Our survey covered a pre- and post-period of 6 months, while in reality, there are mostly several years or decades between exposure and retrospective assessment. Thus, the results of our study might not be transferred directly on the validity of long-term exposure assessment but may give a hint on how the validity of assessment will decrease in time.
The form of questions presented on the duration of knee postures may be critical, as participants had to quote frequency and duration of their postures and were not able to see the result of their total time in knee postures (unless they calculated it for themselves). For that reason, self-reported durations of knee postures even higher than the whole measuring period can be found in both surveys (33.7 % of all data in survey
t
0, 44.5 % in survey
t
1). This effect is also known for other studies using open-ended questions for exposure assessment (e.g. Douwes et al.
2007). As we were only interested in subjects’ assessment behaviour rather than in getting plausible self-reported information, we refrained from excluding implausible data from the analysis as is necessary in an epidemiological study. In order to recognise a possible bias caused by this, we performed a statistical sub-analysis including only data sets from survey
t
0 reporting total duration of knee postures within duration of measuring period. This sub-analysis showed no significant differences relative to the total sample. Furthermore, there were no significant differences in age, profession, education, or number of years in profession between subjects who reported extremely implausible duration of knee postures and subjects giving plausible self-reports.
Taking absolute time units as assessment units (
minutes) may have caused problems, especially for short-term activities. But asking relative percentages of time seemed to be unsuitable as the measuring periods were not of constant duration but had to be applied to particular working situations. Furthermore, there are some hints that subjects may assess the duration of occupational tasks better in terms of absolute time than as percentage of time (Heinrich et al.
2004).
Strengths
The main strength of this study is its examination of self-reports at two different time points to demonstrate the effect of recall bias on the validity of assessment. Most studies on method comparison have only been concerned with short-term validity of self-reports, as done in survey
t
0 of this study. Furthermore, we applied a highly valid and suitable measuring technique as criterion method. In a recent review on method comparison, this kind of reference method is described as being of the highest quality level (Barriera-Viruet et al.
2006).
Both questionnaire and measurement were compared “one to one”, that is, in both surveys, the two methods referred to identical subjects and time periods. Thus, time periods for the self-reports were well defined and matched to the measurement periods, which is also described as a criterion of high quality (Stock et al.
2005; Barrero et al.
2009).
Study samples in survey
t
0 (190 participants) and survey
t
1 (125 participants) must be regarded as large in comparison with related studies. In another recent review, mean sample size of the described ten studies was 104 (SD, 63) or 79 (SD, 30), respectively, for four studies also using measuring techniques as criterion method (Stock et al.
2005).
The additional registration of subjects’ health status allowed the examination of a possible differential misclassification due to knee complaints in assessing work-related knee loading, a relation—as we have found—not yet reported in the literature.