European Journal of Epidemiology 13: 541–545, 1997. 1997 Kluwer Academic Publishers. Printed in the Netherlands.
Parental reporting of childrens’ coughing is biased R.E. Dales1, J. White2, C. Bhumgara3 & E. McMullen4 1
Air Pollution – Health Effects Research Section, Health Canada, and the Department of Medicine and, Epidemiology and Community Health, University of Ottawa, Canada; 2 Building Science, Research Division, Canada Mortgage and Housing Corporation, Canada; 3 Department of Geography, University of Waterloo, Canada; 4 Biostatistics Section, Environmental Health Directorate, Health Canada, Canada Accepted in revised form 29 November 1996
Abstract. Assessment of cough in the clinical setting as well as in community-based studies of respiratory epidemiology has relied on self-reports. To examine the accuracy and potential for systematic bias in reported cough during a field study, questionnaires administered to parents about their childrens’ coughing were compared to overnight cough recordings performed in 145 homes in the community of Wallaceburg, Canada. Percentage agreement between reported and recorded coughing was low, with kappa statistics ranging from 0.02–0.10. Compared to non-smoking parents, smokers under-reported their
childrens’ coughing ( p = 0.01). The association found between parental smoking and recorded coughing was biased towards the null when reported coughing was substituted for recorded coughing: the odds ratio between parental smoking and recorded coughing was 3.1 (95% CI: 1.1–8.8) whereas for reported coughing it was 0.6 (95% CI: 0.2–1.7), the difference in the odds ratios being significant at p = 0.03. When carrying out field surveys, consideration should be given to measuring cough in a subsample of the population in order to estimate the degree of bias inherent in the questionnaire-based results.
Key words: Cough, Epidemiology, Questionnaire, Reporting bias, Respiratory disease, Validation
Introduction Cough is one of the most common respiratory symptoms. It is a manifestation of many diseases and may be the only symptom of asthma [1]. Questionnaire-based field surveys have been used extensively since the 1960s to determine the community prevalence of respiratory symptoms and to provide evidence of adverse respiratory health effects of both indoor and outdoor pollutants [2–7]. The reliability of self-reports is satisfactory [8–11] but agreement between reported and recorded coughing is poor [12–14]. The present study investigates the accuracy with which parents report nocturnal coughing in their children. In a sample of children who participated in an indoor environmental health study, parental reports of nocturnal coughing were compared to the results of overnight cough recordings.
Materials and methods Study group. In the winter of 1993–94 standardized respiratory symptom questionnaires were administered and overnight cough recordings were collected in the community of Wallaceburg Ontario, population approximately 14,000. The study population
used to test the cough recorder was part of a larger study of the health effects of indoor microbiologicals on childrens’ health. A census of elementary schoolchildren in Wallaceburg, Ontario was created using school lists. 1,438 families with school children, identified from school lists, were sent letters of introduction and then telephoned to request participation in the study. The sociodemographics of families with at least one elementary schoolchild in this region of the province were as follows: the average family size was 4.4 per household, 59% of family incomes were above Can$ 50,000, at least one parent had completed secondary school in 76% of families, and 88% of homes were detached single family dwellings [15]. The questionnaire. A trained interviewer administered the questionnaire to the parent or guardian. The questionnaire had been previously standardized, used in this population, and tested for reproducibility [16]. Using the test-retest method, percentage agreement for the report of persistent cough was 95% with a kappa statistic of 63%, indicating moderate reliability [17]. Two of the cough questions were taken from the ATS-DLD standardized respiratory questionnaire [18]. These were: ‘Does this child usually cough?’, and ‘Does this child usually cough during the night or first thing in the morning?’. Two supplemental questions, administered the morning following the
542 cough recording, were ‘How was the child’s cough overnight?’ and ‘How was the child’s cough during the first hour after awakening?’ Possible answers included none, mild, moderate, and severe. The cough recorder. The equipment used consisted of a 486 notebook computer attached to a directional microphone. Data was collected at 4,000 Hz, directly recorded in digital format, and saved as two channels. The first channel recorded noise at 4,000 Hz only when the volume was above 125% of the base line averaged over 10 sec. The second channel resampled the maximum of every 40 data points received at 100 Hz. Cough was defined from the first channel as a sharp burst of sound with an explosive peak such that the slope between the onset and the maximum amplitude was greater than 20 microvolts (µv)/sec. To eliminate false positive results from clicks, static, sharp bangs, and talking, the event had to be at least 0.055 sec duration, and at least 5% of the samples in each event had to have a frequency between 750 and 2,000 Hz. All noises from a recording session were captured, and diagnosed as coughs or not by the computer. The output was number of coughs, duration of coughing, and time of recording. To validate the ability of the computer to distinguish coughing from other extraneous noises, the computer recordings could be played back and listened to at 4,000 Hz. To test the reliability of the listeners, two research assistants independently assessed 119 acoustic sounds (e.g. coughing, talking, walking, doors closing) generated by a convenience sample of volunteers. Agreement on the classification of cough vs other noises was 95%. The instrument was then tested under controlled circumstances in two hospital sleep laboratories. Of the subjects monitored overnight, events occurred in 12 patients. Of 280 coughs identified by the judges, the cough recorder detected 245 resulting in a sensitivity of 88%. Of 262 coughs detected by the cough recorder, the judges agreed in 245 cases giving a positive predictive value of 94%. There were 17 false positive coughs due to other noises. A pilot study of 8 children was then conducted in Ottawa, Ontario homes. Doctors in clinics were canvassed for children and adults with a history of either acute or chronic cough. When the microphone was positioned within one metre of the subject in a reasonably quiet room (without excessive talking or a television set), the cough recorder captured 100% of coughing events as judged by a research assistant listening to the recordings. During the field study, the cough recorders were set up with the directional microphone approximately 1 meter from the child’s pillow. A low intensity cough generated by the field technician was used to set the gain of the cough recorder. At night the participant’s parents were asked to turn on the cough recorder when the child retired to bed. The computer automatically shut off after 10 hours of recording.
Results Of 403 children for whom a completed respiratory symptom questionnaire was available, acceptable cough recordings were available for a nonselected subsample of 145. Those with (n = 145) and without (n = 258) available cough recordings are contrasted in Table 1. There were no systematic differences in variables which may influence the number of coughs recorded and the reporting of coughs. The average number of nocturnal coughs per subject was 12 with a range of 0 to 234. The average number of coughs per hour was 1.3 with a range 0 to 30, while the corresponding values for mean duration (range) were 0.2 sec/hour and 0 to 4.7 sec/hour. Validation of the cough recordings obtained in the field was performed by listening to fifteen randomly chosen sounds from each of 25 randomly chosen children (Table 2). Of 83 coughs heard by the listener, the computer correctly diagnosed 66, giving a sensitivity of 80%. Very quiet coughs were occasionally missed. Of 186 noises not thought to be coughs by the listener, the computer incorrectly classified 9 as Table 1. Characteristics of the index child and the parent completing the questionnaire stratified by participants (those with cough recordings) and non-participants (those without cough recordings), Wallaceburg, Canada, 1993– 1994 Characteristic
Index child Age, mean (SD) Gender, female Asthma Usual cough
Participants (n = 145)
Non-participants (n = 257)
009.6 (2.7) 077 (53%) 034 (23%) 049 (34%)
009.3 (2.7) 127 (49%) 056 (22%) 070 (27%)
Parent completing the questionnaire Gender, female 103 Education less than secondary 073 Income less than Can$ 50,000 057 Asthma 019
(71%)
177 (69%)
(50%)
146 (57%)
(39%) (13%)
103 (40%) 031 (12%)
Table 2. The ability of the computer to distinguish a cough from extraneous noises was estimated by the judgement of someone listening to the overnight acoustic recordings in Wallaceburg children, 1993–1994 Computer identification Human judgement
Cough present Cough absent (n = 75) (n = 197)
Cough present (n = 83) Cough absent (n = 189)
66 09
017 180
543 coughs, giving a false positive rate of 5% or a specificity of 95%. Overall accuracy was 90%. In the field study, the mean and median values for duration of coughing per hour were two-threefold higher when parents reported coughing (Table 3). However, due to the large random variation in reported coughing, evidenced by the large standard deviations of the means, differences in recorded coughing between children reported and not reported to be coughing, were not statistically significant at p < 0.05. Misclassification of subjects with cough was very high, evidenced by the low percentage agreements, less than 50%, and kappa statistics less than 0.10. Having the objective measure of recorded cough allowed the determination of factors influencing reported coughing (Table 4). Reported coughing was much less sensitive than recorded coughing. Compared to non-smokers, parents who smoked underreported coughing in their children ( p = 0.01). Children of smokers had the highest prevalence of
recorded coughing but the lowest prevalence of reported coughing. The effect of this reporting bias on observed associations between cough and risk factors for respiratory disease was evaluated (Table 5). A positive association between parental smoking and childrens’ coughing was detected by cough recordings (odds ratio: 3.1; 95% CI: 1.1–8.8). This association was biased towards the null when reported coughing was substituted (OR: 0.6; 95% CI: 0.2–1.7).
Discussion Most cough recorders described in the medical literature consist of overnight tape recordings from which a listener subsequently counts the coughs over a specified period of time [12, 19–21]. The sample sizes were relatively small with one exception [14], and the repeatability of the listener’s score was
Table 3. Measured duration of coughing per hour stratified by parental reporting of child’s cough. Wallaceburg, Canada, 1993–1994 Parental report
No. with coughing
Agreement % (kappa)
Cough duration (sec/hour) Median
75th percentile
Usually coughs No (n = 95) Yes (n = 49)
64 39
49% (0.10)
0.03 0.10
0.17 0.22
Usually coughs at night or early in the morning No (n = 86) Yes (n = 47)
57 37
50% (0.10)
0.03 0.06
0.14 0.23
Coughed the previous night No (n = 107) Yes (n = 31)
76 23
39% (0.02)
0.03 0.13
0.17 0.28
Coughed in the morning No (n = 104) Yes (n = 38)
73 30
43% (0.06)
0.04 0.09
0.17 0.24
Table 4. The influence of selected risk factors for respiratory disease on reported cough. Using recorded coughing as the criterion, reported cough was significantly biased by parental smoking. Wallaceburg, Canada, 1993–1994 Risk factor for cough
Prevalence of nocturnal cough Recorded
Reported
Difference
Parental asthma
– present (n = 18) – absent (n = 117)
83% 69%
22% 21%
61% 48%
Household income
< Can$ 50,000 (n = 55) > Can$ 50,000 (n = 55)
78% 64%
20% 24%
58% 40%
Parental smoking
– present (n = 37) – absent (n = 101)
86% 66%
16% 24%
70%* 42%
* p = 0.01 using a chi-squared test.
544 Table 5. Differential misclassification of reported coughing using recorded coughing as the criterion stratified by selected risk factors for respiratory disease: parental history of asthma, household income, and parental smoking. Wallaceburg, Canada, 1993–94 Risk factor for cough
Prevalence of nocturnal cough Reported
Recorded
Present
Absent
OR (95% CI)
Present
Absent
OR (95% CI)
Parental asthma
– present – absent
04 25
14 92
1.1 (0.3, 3.5)
15 81
03 36
2.2 (0.6, 8.2)
Household income
< Can$ 50,000 > Can$ 50,000
11 13
44 42
0.8 (0.3, 2.0)
43 35
12 20
2.1 (0.9, 4.8)
Parental smoking
– present – absent
06 24
31 77
0.6 (0.2, 1.7)
31 67
05 34
3.1 (1.1, 8.8)*
* p = 0.03, two-tailed t-test of the difference in the logs of the odds ratios.
generally not formally tested. There was poor agreement between measured and self-reported cough frequency. In the present study, automated analysis facilitated larger sample sizes by eliminating the need to listen to the entire recording, and agreement was good between the computerized scoring and scoring by the listener. Although our sample size was much larger, it remains a convenience sample. The overall prevalences of symptoms and cough may not be representative of the general population, but this should not have influenced the internal validity of our findings that cough reports are relatively inaccurate and more importantly, biased. Although this study cannot determine how much coughing is normal, it demonstrates that the answer to this will depend on whether it is a parental report or a measurement. If a parental report, it will also depend on his/her characteristics. Parental reporting of cough is inaccurate. The discrepancy between reported and observed coughs is unlikely due to error in the recordings which demonstrated a high degree of accuracy both in a sleep laboratory and in the field. The sensitivity of the recorder was calibrated for each case, and even if relatively insensitive, it would not have reduced the relationship between measured and reported coughs but simply reduced the total number of coughs detected. Therefore it is likely that parental reporting of overnight coughing is inaccurate. Parents may not hear their children cough at night because the distance between bedrooms may be too great, and because the parents may not have been awakened by the coughing. One of the cough questions, ‘Does your child usually cough?’, which does not specify a particular time of day, may not accurately reflect nighttime cough if parents are more aware of daytime coughing and daytime coughing has different determinants than nocturnal coughing. Daytime coughing due to asthma may be provoked by physical exertion
or exposure to the cold winter air, whereas post nasal drip, esophageal reflux, and dust mite allergy may be more likely to provoke nocturnal coughing [22]. Reported cough can bias estimates of symptomexposure associations. This study provides evidence that relying on reported cough may systematically bias the estimated effects of selected risk factors for respiratory disease: parental history of asthma (a familial risk factor), household income (a socioeconomic factor), and tobacco smoke (an environmental factor). Odds ratios between these risk factors and reported cough were 1.1, 0.8, and 0.6 respectively whereas with recorded cough the odds ratios were 2.2, 2.1, and 3.1, the difference being statistically significant only for smoking. This demonstrated bias suggests that previous questionnaire studies of tobacco smoke–health effects may have underestimated the true magnitude of effect. In summary, parental reporting coughing in their children is not accurate, with a large amount of random error and evidence of systematic bias. These findings should be taken into consideration when interpreting respiratory symptom questionnaires. Cough recordings are an accurate indicator of nocturnal coughing in children, providing quantitative measures, and are free from the reporting biases which may influence questionnaires. In future field surveys, consideration should be given to objectively quantifying cough in a subsample in order to quantify the bias in results based on questionnaire reports of cough.
Acknowledgments This work was supported by the Panel for Energy Research and Development (PERD), Natural Resources Canada, and Canada Mortgage and Housing Corporation, Canada. The authors wish to
545 acknowledge Melville Corporation who developed the cough recorder and tested it under laboratory conditions, and Magellan Engineering who carried out the field study.
References 01. Fuller RW, Jackson DM. Physiology and treatment of cough. Thorax 1990; 45: 425–430. 02. Dodge RP, Burrows B. The prevalence and incidence of asthma and asthma-like symptoms in a general population sample. Am Rev Respir Dis 1980; 122: 567–575. 03. Burrows B, Lebowitz MD. Characteristics of chronic bronchitis in a warm, dry region. Am Rev Respir Dis 1975; 112: 365–370. 04. Schenker MB, Samet JM, Speizer FE. Risk factors for childhood respiratory disease. Am Rev Respir Dis 1983; 128: 1038–1043. 05. Dales RE, Raizenne M, El-Saadany S, Brook J, Burnett R. Prevalence of childhood asthma across Canada. International J Epidemiol 1994; 23: 775–781. 06. Abramson M, Voigt T. Ambient air pollution and respiratory disease. Med J Australia 1991; 154: 543–553. 07. Dockery DW, Speizer FE, Stram DO, Ware JH, Spengler JD, Ferris BG Jr. Effects of inhalable particles on respiratory health of children. Am Rev Respir Dis 1989; 139: 587–594. 08. Lebowitz MD, Burrows B. Comparison of questionnaires: The BMRC and NHLI respiratory questionnaires and a new self-completion questionnaire. Am Rev Respir Dis 1976; 113: 627–635. 09. Comstock GW, Tockman MS, Helsing KJ, Hennesy KM. Standardized respiratory questionnaires: Comparison of the old with the new. Am Rev Respir Dis 1979; 119: 45–53. 10. Helsing KJ, Comstock GW, Speizer FE, Ferris BG, Lebowitz MD, Tockman MS, Burrows B. Comparison of three standardized questionnaires on respiratory symptoms. Am Rev Respir Dis 1979; 120: 1221–1229. 11. Withey CH, Price CE, Swan AV, Papacosta AO, Hensley MJ. Repeatability of a questionnaire to assess respiratory symptoms in smokers. J Epidemiol Community Health 1988; 42: 54–59.
12. Falconer A, Oldman C, Helms P. Poor agreement between reported and recorded nocturnal cough in asthma. Pediatr Pulmonol 1993; 15: 209–211. 13. Hsu JY, Stone RA, Logan-Sinclair RB, Worsdell M, Busst CM, Chung KF. Coughing frequency in patients with persistent cough: Assessment using a 24 hour ambulatory recorder. Eur Respir J 1994; 7: 1246– 1253. 14. Brooke AM, Lambert PC, Burton PR, Clarke C, Luyt DK, Simpson H. Night cough in a population-based sample of children: Characteristics, relation to symptoms and associations with measures of asthma severity. Eur Respir J 1996; 9: 65–71. 15. Ontario Health Survey 1990 User’s Guide, Volume 1: Documentation. Toronto, Ontario: The OHS 1990 Information, Planning and Evaluation Branch, Ministry of Health. 16. Dales, RE, Schweitzer I, Bartlett S, Raizenne M, Burnett R. Indoor air quality in health: Reproducibility of respiratory symptoms and report at home dampness and molds using a self-administered questionnaire. Indoor Air 1994; 4: 2–7. 17. Fleiss JL. Statistical methods for rates and proportions, 2nd edn. New York: John Wiley and Sons, 1981. 18. Ferris BG Sr. American Thoracic Society Executive Committee, Epidemiology Standardization Project. Am Rev Respir Dis 1978; 118: 1–120. 19. Woolf CR, Rosenberg A. Objective assessment of cough suppressants under clinical conditions using a tape recorder system. Thorax 1964; 19: 125–130. 20. Loudon RG, Brown LC. Cough frequency in patients with respiratory disease. Am Rev Respir Dis 1967; 96: 1137–1143. 21. Power JT, Stewart IC, Connaughton JJ, Brash HM, Shapiro CM, Flenley DC, Douglas NJ. Nocturnal cough in patients with chronic bronchitis and emphysema. Am Rev Respir Dis 1984; 130: 999–1001. 22. Smolensky MH, D’Alonzo GE. Medical chronobiology: Concepts and applications. Am Rev Respir Dis 1993; 147: S2–S19.
Address for correspondence: Robert Dales, MD, Ottawa General Hospital, Room LM-17, 501 Smyth Road, Ottawa, Ontario, Canada K1H 8L6 Phone: (613) 737-8198; Fax: (613) 737-8141