Effect of masking on circadian adjustment and interindividual differences on a rapidly rotating shift schedule.

OBJECTIVES
The aim of the present study was to define the effect of masking on the estimation of circadian adjustment of and interindividual differences between nurses on a rapidly rotating shift schedule.


METHODS
The phase shift of the circadian rhythm of rectal temperature was studied in 17 subjects. The following three different methods for estimating the circadian phase shift were compared: (i) cosinor analysis, (ii) a method using normative endogenous data in which the masking effects are removed ("purification"), and (iii) a method using normative endogenous data in which the masking effects are not removed.


RESULTS
The mean phase delay during the second night shift (compared with the morning shift) was 6.3 h according to the cosinor analyses of the raw data and 6.2 h according to the cross-correlation method of the raw data. When the masking effects were removed ("purification"), the phase shift was only 1.7 h. Diurnal type did not significantly explain the differences between individuals in the amount of phase shift of the raw data but was the only significant factor, explaining 32% of the variation between individuals in the phase shift of the purified data.


CONCLUSIONS
Masking effects on body temperature should be taken into account before any definite conclusions can be drawn about the relation between individual factors and the adjustment of the circadian rhythm of body temperature.

During nigh t shifts the circadi an " profi le" of the masking effect s of sleep and wor k are dramatically cha nged compared wi th that of morning or evening shift s. Wh ile the order of the main dail y ac tivities during morning shifts is sleep, work, leisure, during night shifts the order is mostly sleep, leisure, work. As has been shown recentl y in a field study of night work on a slow ly rotating shift syste m (6), the circa di an adju stm ent between morning and night shifts is con siderabl y overes timated if masking effects ca use d by the enviro nme nt and life -style on bod y temperature are not rem oved.
Circadian adj ustment to shift work has been reported to be dependent on interindividual differences su ch as morningness-eveningness ( I ), introver sionext ro version (7 ), neu roticism (8), and shift-work toler an ce (I , 3, 4), although contradic tory re sults hav e al so been publi shed (9,10). Since the ci rca dian rhythm of bod y temperature du rin g shift work is greatly influenced by masking, the question aris es of wh ether the interindividual differences rep orted to explain circadian adjustment are ac tually explaining the endog e no us or onl y the exogenou s infl ue nces up on bod y temperature.
Th e obj ectives of the present study were to detine the effect of ma sk ing on the estimation of circa dia n adjustment to a rapidly rot at ing shift sc hedule and to study the effec t of mask ing on the relation between interindividual differences and circadian adj ustme nt.

55
Scand J Work Environ Health 1994, vol 20, no I

Subje cts and shift schedule
After a preliminary questionnaire, 20 women on a rapidly rotating shift schedule and one woman working permanently in night work volunteered for the study. All of the subjects had at least 1.5 years' experience in shift work and were either nurses or nursing aides in a hospital in Kuopio , Finland. From the group working on the rapidl y rotating shift schedule, rectal temperature was successfully recorded for 18 subjects. One of these subjects was also excluded since she showed no clear rhythmicity during the second night shift (the percentage rhythm < 10%), which precluded useful analysis . The characteristics of the 17 subjects whose data were analyzed are given in table I.
The subjects worked a rapidly rotating three-shift system with morning, evening, and night shifts irregularly placed in the shift cycle of three weeks. Morning shifts (0700-1500) and evening shifts (1300-2100) lasted 8 h and the night shifts (2100-0700) 10 h. There was an average of 38 h of work per week, and one shift cycle (three weeks) included an average of seven morning shifts, five evening shifts, and three night shifts. The night shifts occurred mainly as one or two nights in succession. No measurements were accepted during the expected ovulation day and during menstruation, information on which was requested from every subject. the masking effects of sleep, work, travel, and strenuous activity.
The effect of masking was analyzed mainly by comparing the shifts assessed by method s 2 and 3. Since many earlier studies have used cosinor analy sis of the raw data, this method was also carried out to make comparisons with earlier results possible.
The subjects performed their normal daily routines during the study. The body temperature measurements started at 0700 at the beginning of the first morning shift and ended soon after the second night shift at about 2000. The morning shift followed another morning shift or a free day, but never an evening or a night shift.
Body temperature and activity recordin g Body temperature was measured every 5 min by a rectal probe inserted 10 ern beyond the external anal sphincter. The accuracy of the measurement device was 0.01°C (11). (The measuring device s were calibrated by an extreme precise quartz-thermometer of the accuracy of 0.001QC). Yellosprings YSI 400 thermistors were used. No subject was studied on the predicted day of ovulation or during menstruation. The subjects kept a daily record of the times of the following activities: sleep, leisure, travel, work, and strenuous activity. St udy design Circadian adjustment to night work was estimated from calculations of the phase difference of the body core temperature rhythm from the last morning shift to the second night shift. The following three different methods were used for the phase estimation of the circadian rhythm of body temperature: (i) cosinor analysis of the "raw" (ie, unchanged) data, (ii) analysis of the "raw" data through comparison with normative endogenous data (see the section entitled "Normative Endogenous Data and Purification" and references 5 and 6), and (iii) analysis of the data with the second method but after "purification" to remove a Mornlngness, minimum + 1, maximum + 6. Estimation of the circadian rhythm of body temperatu re Cosinor analysis . Cosinor analyses of the raw data were performed according to the single cosinor method ( 12). The analyses were performed upon hourly data obtained from two 24-h windows start ing at about 0800. Hourly data have been presented instead of data measured every 5 min so that exactly the same data format can be used for the three different methods of estimating the circadian phase . However, cosinor acrophases were similar whether the 5min or the hourly data were used, with no significant differences between the acroph ases.
The first window started at the beginning of the second morning shift, and the second window covered the interval between the end of the first and second night shifts. The phase shift of the rhythm was estimated from calculations of the difference in acrophases between the two windows for each subject.
Norma tive endoge nous data and purificati on. The method of comparing the "raw" data with normative endogenous data requires hourly reading s of rectal temperature, and it has been described in detail elsewhere (5,6). It is based upon the concept of "purification" as used by Wever (13).
The basic concept is that a measured rhythm consists of exogenous and endogenou s components that 37 (sleep length and quality) , A detailed description of the index has been published earlier (16). Marital status was classified as subjects living alone or living with a partner. Maximal oxygen consumption (Y0 2 max) was predicted by two different bicycle ergometer tests according to the method recommended by the World Health Organization (17). Real oxygen consumption (Oxycon-4, Mijnhardt, Holland) and the average maximal heart rate for the subjects' age group were used in the prediction,

Results
The mean circadian phases obtained with the different methods are summarized in Leisure Sleep act additively. The shape and timing of the exogenous component depend upon several factors, including, mainly, the individual's sleep and activity cycle, sleep lowering the body temperature and different types of activity raising it by different amounts. The endogenous component, whose phase is sought, is assumed to be described by "normative endogenous data." These data are the mean rhythm for the rectal temperature of more than 50 normal subjects (18-25 years of age) studied while awake and in the supine position. Under these conditions, masking effects have been minimized. The method requires a record of the main type of activity in the previous hour (the choices being, eg, asleep , lying, sitting , standing, walking). The method then "purifies" the observed temperature data according to the following rules: We then tested all combinations of A through D and also, for each combination, shifted the "purified" rhythm by 0-23 h in hourly steps. The combination of shift and values for parameters A through 0 which gives the best match with normative endogenous data (ie, minimizes the summed squared deviations between the purified and normative endogenous data) is taken as indicating the shift of the endogenous component in the original data together with the sizes of the masking effects. Figure I gives an example of how the purification worked for one subject. Such an analysis gives (i) a measure of the shift of unchanged (raw) data, (ii) a measure of the shift of changed (purified) data, and (iii) estimates of the masking effects due to sleep, leisure, travel, and work.

lnterindividual differences
The interindividual differences between the subjects were studied with a questionnaire (age, diurnal type, neuroticism, shift-work tolerance, extroversion , marital status, and number of children) and by laboratory measurements of weight, height, and maximal oxygen consumption.
Diurnal type was estimated by the "diurnal type scale" according to Torsvall & Akerstedt (14). Diurnal type (morningness-eveningness) correlates with the morning and evening disposition of the sleepwakefulness, as well as with other circadian rhythms of the subjects. Neuroticism and introversion-extroversion were assessed according to Eysenck (15). Shift-work tolerance was estimated with the use of a questionnaire using three criteria: the amount of general fatigue during the last three weeks, the prevalence of digestive disorders , and sleep disturbance 57 Scand J Work Environ Health 1994, vol 20, no I Table 2. Circadian phase (decimal time of the minimum) of body temperature during the days including the morning and the second night shifts, the change from the morning to the second night sh ift, phases fr om the cos inor analys is and from the method using normative endogenous data using both raw and purified data, and the stat isti cal sign if icance of the phase shift rnorninq-second night shift.  cant when the raw data were used (P < O.OOOI) but insignifica nt (P = 0.1050 , two-tail ed) when purified data were used.
These results were almost identical if the cosinor analysis was performed upon the 24 hourly points for each day (the same time points as used with the methods for normative endogenous data ) rather than the 288 5-min values. Cosin or anal ysis also indicated that there was a significant fall of ampl itude in the raw data durin g the second night shift (P< O.OI) . Figure 2 shows the distribution of the phase minimums in relation to midsleep (the midp oint of the times of falling asleep and wakin g up) durin g the M and 2N shifts. The midsleeps shifted slightly more than the phases of the raw data.
58 lnterindi vidual differences A step wise regression analysis was performed to study the dependence of the "raw" and "purifie d" phase shifts on the factors age, weight, height, diurnal type , neurotici sm , shift-work toler ance, introversion-extroversion, and physical fitness (maximal oxygen co nsumption). Since the regr ession analysis with both anal ytic al meth ods using raw data produced the same results, onl y the results of the two methods using normative endogenous data (with raw and purified data) have been reported.
The factors explained 61% of the variation of the "raw" phase shift s (table 3). Neurotici sm explained 18% of the variati on in the phase shifts and had the greatest predictive power, together with height. Neu-• • Figure 3. Cor rela tion (r =0.56, P <O.019 ) between diurnal type (4 =extreme morningness, 1 =extreme eveningness) and the phase shift of the pur ified data between the morning and second night shift. Table 3. Stepw ise regression analyses of the phase shift of the body tempe rature from the morn ing to th e seco nd night shift with the use of the c ross-correlat ion methods with t he raw and purif ied data. The studied individual factors expla in ed 61% of the variat ion of the phase shift of the raw data (R-square 0.61, total df 16 Since the main difference between this study and that of Minors & Waterhouse (6) appeared to be in the speed of rotation of the shift schedule, it seems possible that subjects on rapidly rotating shift schedules adjust their endogenous rhythmi city slower than do subjects with longer spans of night work. It is possible that a more rapid circadia n adj ustment may be beneficial only in slow ly rotating shift systems, be-

Discussion
The circa dian phase delay of the pur ified data was 4.6 h less than the phase delay of the raw data. The change in the cosinor acroph ase was also very similar to the pha se dela y of the raw data assessed by the method using normative endogenous data . Our study confirms the findings of Minors & Waterhou se (6) that conventional meth ods which do not remove the masking effects of body temperature greatly overestimate the amount of circadian adju stment during night shifts. Th e prob able explanation is that sleeping and waking were delayed on the night shift, coupled with the fact that most activit y took place (during the nigh t work itself) in the seco nd half of the waking span rather than in the first half on the morning shifts. As a further consequence there would also be a lack of phase coincidence between the endo genous and exogeno us components of the circadian rhythm, and this lack could account for the observed decre ase in rhythm amplitude during the night shift.
The phase delay of the purified data was only 1.7 h during the two night shifts. In agreement with this finding, earlier studies based on oral temperature data had already indicated that , in rap idly rotating threeshift work (1-2 nights in succ ess ion), the adjustment of the tempera ture rhythm was only marginal ( 18,19). It is surprising ther efore that , in a recent study of nurses workin g a greater numb er of successive night shifts (6), a similar methodology was used (nur ses, conti nuous rectal temperatur e recording, maskin g effects remove d), but the average phase delay during the same time period (two nights) was clea rly greater, 4.6 h. roti cism was not associated with a phase delay in the raw data , the correlation being -0.43 (P < O.I). The correlation between neur oticism and the phase shifts in the cosinor analysis was -0.42 (P <0.1 ).
The interindividual factors expl ained somewhat less, 46%, of the varia tion in the phase shifts of the purified data. Diurnal type expl ained 32% of the variation and was the most important and also the only signific ant factor in the regression model. Even ingnes s was associated with phase delay ( cause in rapidly rotating shift schedules circadian adju stm ent may be unnecessary or even detrimental -causin g internal desyn chronization and a decrease in shift-w ork tolerance. Therefore, some differences might exist bet ween the subj ec ts working these different schedules . In part these differ ences may be a form of natural selection, subjects who innately adju st more slow ly being more suited to rapidly rotating shifts and, ther efore , more likel y to stay in them . But , in addition, there might be behavioral differe nces , thos e in slowly rotating shift systems adj usting their life-styles to a greate r extent. That is, "motiva tion" to adju st might be more appropriate onl y with a slow rotation of shifts; by contrast, a better cour se of action for a subject on a rapidl y rotating shift system would be to atte mpt to maintain a stable phasing of circadian rhythms (and one that is phased appropriately for a diurnal life-style) as much as possible (20, 2 1).
Diurnal type did not sig nificantly expl ain the variati ons in the pha se shifts of the raw data, but it was the only sign ificant factor, ex plaining 32 % of the varia tion of the phase shifts of the purified dat a. Morn ing "types" had a tendency towards slower ci rcadia n adj ustme nt. It has been hypoth esized that morningness is linked to a shorter end ogen ous period of circa dian rhythm icity (9,22). If true , this hypothesis would expl ain why morning type s are not able to phase del ay their circadian rhythm s during night shifts as well as evenin g typ es. Although earlier fie ld studies, which have not removed the masking effects of the circadian rhythm of body temperature, have found a similar rel ation between morningness and phase delay (l, 23, 24), the interference due to maskin g effec ts might explai n so me earlier negative findings . (See, eg, referen ces 9 and 25 .) Our results suggest that the observed differences between morning and evening types in their ability to adjust to night work are not easily explained in term s of differences in beha vior bein g initially responsible.
The degree of neuroticism was a sign ificant predictor of the interindividual difference s in the phase shift of the raw dat a, but it did not explai n the phase shifts when the masking effect s were remo ved . Vidacek et al (25 ) have shown that, although neuroticism correlates highl y with the frequency of self-reported symptoms in shift work, in pro spect ive studies, a person' s neuroticism before starti ng shift work does not predict future shift-work tolerance. This finding demonstra ted that poo r shi ft-w ork tol eran ce might induce neuroticism. Since poor shift-wo rk tolerance mean s, in most cas es, sleep disturbances, the masking effect s of sleep or other life-styl e differenc es could explain the observed relation ship between neuroticism and faster adjustment of the raw, unpurified dat a.
Shift-work tolerance correlated highly with neuroticism, and it is probable that for this reas on it was not selected into the regre ssion mod el which expla ined the variation of the phase del ays of the raw 60 data. Shift-work tolerance corre lated, however, with the phase delay of the raw data . Since shift-work tolerance did not correlate with the phase shift of the puri fied data, it can be spec ulated that the earlier observa tions of faster circadian adju stment of subjects with poor shift-work toleranc e (l, 3, 4) co uld again be due to the use of data from which ma skin g effec ts on the circadi an rhythm of bod y temperature had not been rem oved .
It seems clear that in future fiel d studies of the circadian rhythm of body temperature, the maskin g effect s on body temperature should be taken into account or, preferably, remove d to increase the value of the results . In particul ar, the masking effec ts on body temperature should be remo ved before any definit e co nclusions about changes in the endoge nous component of temperature circadian rhythmicity can be drawn.