Non-Response in Telephone Surveys of Copd Patients Does Not Introduce Bias

Abstract

We examined the telephone response rates in a 2-year study of patients with COPD to identify factors associated with non-response. A total of 381 patients received monthly telephone calls to assess symptoms and treatment in 2007 and 2008. A total of 9019 calls were made over 24 months, of which 73% were answered. The highest response rate was obtained in February 2007 (81 %), and the lowest in July 2008 (48%). The monthly response rate was lower in the second year of follow up. There were 39 patients (10%) who were less frequent responders, with 10 or fewer responses. Less frequent responders were more likely to be current smokers and have hypoxaemia. The median number of answered calls was 18 (interquartile range 16–20). In bivariate models the median number of responses was significantly lower in subjects with chronic cough. In multivariate models neither demographic nor disease characteristics were significantly associated with non-response. We found less frequent responders to monthly telephone calls to be similar in characteristics to frequent responders. This suggests that non-response does not necessarily introduce bias in telephone surveys of patients with COPD.

Introduction

Telephone surveys have been used in chronic obstructive pulmonary disease (COPD) research to assess the frequency of exacerbation, the quality of health care, the burden of disease and the cost of health care.^1–7 Response rates in telephone surveys vary greatly.^1,5,6 If non-response is associated with demographic characteristics or clinical features that also are associated with the study outcomes, then subsequent analyses will be biased. Potential bias due to non-response in postal surveys has been described earlier^8–11, but to our knowledge, no study has assessed how non-response to telephone calls in studies of COPD patients might influence the inferences drawn from such data. To evaluate both the internal and external validity of a telephone survey, it is crucial to know whether the frequent responders to telephone calls have the same characteristics as non-frequent responders.

Based on previous studies of non-responders in respiratory epidemiological studies^8,9,12, we hypothesised that men, current smokers and less well-educated people would be less inclined to respond to regular telephone calls regarding exacerbation in COPD. We have examined the demographic and disease characteristics associated with non-frequent response in a study of 381 COPD patients from Western Norway receiving monthly telephone calls for 2 years. The research questions were: (1) does the response rate to monthly telephone calls change over two years? (2) are there any demographic details associated with response to monthly telephone calls in COPD?

Methods

During 2006 patients were recruited through outpatient clinics from several hospitals in Western Norway, and from three private specialist practices in Bergen. All patients had a smoking history of more than 10 pack-years, a post-bronchodilator FEV₁/FVC ratio < 0.7, and an FEV₁ of < 80% predicted.¹³ Patients with known cancer the last five years, or inflammatory disorders like rheumatoid arthritis, systemic lupus erythematosus or inflammatory bowel disease were not included. Baseline characteristics such as smoking habits, information on symptoms and prior acute exacerbations of COPD, and measurements of lung function and arterial blood gas were obtained at the first visit to the study centre. Details of the study sampling and inclusion criteria may be found elsewhere.¹⁴ Written information was provided and written consent was obtained from each participant prior to inclusion. The study was approved by the appropriate ethics committee. Before the telephone survey started in 2007, 6 patients died, 16 were excluded or withdrew. Also, the last 22 patients that were recruited were not included in the telephone survey.

The telephone survey was conducted in 2007 and 2008. Patients were called once a month by trained staff (medical students). If there was no answer, or the patient did not have the time to be interviewed, he or she received a second call within a week. Interviews followed standardized forms, addressed symptoms and changes in treatment, and lasted approximately 2–3 min. Dyspnoea was assessed by the Medical Research Council breathlessness scale, which defines dyspnoea grade 2 as breathlessness when hurrying on the level or walking up a slight hill, grade 3 as walking slower than most people on the level or stopping after 15 min when walking at own pace. Patients who died, were excluded or did not want to participate in interviews any more during follow up, were not called. These values were recorded as missing in the data analyses.

Analyses were performed using a standard package (Stata, StataCorp, College Station, TX, USA). Logistic regression using Generalized Estimating Equations (GEE) was used to investigate the factors associated with repeated response to calls. Patients who dropped out were included in the analyses until date of death or exclusion. Two models were fitted: one including only demographic variables and the other using disease characteristics in addition to demographic factors as explanatory variables. In both models the dependent variable was repeated telephone calls.

Results

Of the 390 subjects included in the telephone survey, 9 patients left the study (8 died, 1 excluded) during the first months of 2007, leaving 381 subjects. The response rate for each month is shown in Figure 1. The total number of calls during the 24 month study period was 9019, of which 73% were answered. The highest response rate was obtained in February 2007 (81%), and the lowest in July 2008 (48%). There was a slight decrease in monthly response during the study period. In the first 12 months of the study the patients responded to 75% of the calls, while in the last 12 months the corresponding value was 69%. The median number of responded telephone calls was 18 (interquartile range 16, 20).

Figure 1

Response to telephone calls in 2007 and 2008. There were 381 persons in the study and all were called every month (except in cases of withdrawal from the study or death)

The distribution of answered calls is shown in Figure 2. When separating between less frequent and frequent responders, using the 10th percentile of response as cut off, the less frequent responders (≤10 received calls) were more likely to be current smokers and have hypoxaemia (PaO₂ < 8.0 kPa) (Table 1). The median number of responses varied significantly with presence of chronic cough; subjects with chronic cough responded less frequently than those without (P = 0.029, Table 2). In multivariate analyses, the association of non-response and current smoking was not significant (Table 3). In a similar model taking disease characteristics into account, there was no association between chronic cough and non-response (Table 3). In the two multivariate models with adjustments for demographic variables and disease characteristics, only month number was significantly associated with response rate (P < 0.001).

Figure 2

Distribution of unanswered telephone calls in 2007 and 2008

Table 1

Characteristics of the study population. Significance was tested with chi-squared for categorical variables, students t-test for continuous variables with a normal distribution (FEV₁ and pack-years), and Mann-Whitney U and Kruskal Wallis tests for continuous variables with skewed distributions (age)

	Responded to ten or fewer telephone calls (n = 39)		Responded to more than ten telephone calls (n = 342)		P-value^*
Sex
No of women	16	41%	134	39%	ns
No of men	23	59%	208	61%	-
Age
Median age, years	62	IQR 57–69	64	IQR 59–69	ns
No aged < 55	6	15%	37	11%	ns
No aged 55–65	18	46%	160	47%	-
No aged > 65	15	39%	145	42%	-
Smoking status
Mean duration, pack-years	42	SD 19	41	SD 23	ns
No of ex-smokers	14	37%	203	59%	0.008
No of current smokers	24	63%	139	41%	-
Education
No with primary education	8	22%	74	23%	ns
No with secondary education	25	68%	212	66%	-
No with university education	4	11%	34	11%	-
Civil status
No married or cohabitating	25	64%	254	74%	ns
No divorced or single	14	36%	88	26%	-
Lung function
FEV₁, L	1.47	SD 0.5	1.51	SD 0.5	-
GOLD stage
II	16	42%	161	47%	ns
III	20	53%	144	42%	-
IV	2	5%	37	11%	-
Chronic cough
No	20	53%	152	46%	ns
Sputum
No	26	68%	196	58%	ns
PaO₂
No with ≥ 8 kPa	26	77%	276	89%	0.046
No with < 8 kPa	8	24%	36	12%	-
Dyspnoea
No with grade 2	34	90%	261	81%	ns
No with grade 3	19	51%	137	42%	ns
No of exacerbations in the previous year
0	25	66%	203	59%	ns
1 or 2	10	26%	112	33%	;
≥3	3	8%	27	8%	-

ns, P ≥ 0.05

Table 2

Number of telephone calls answered

	Median	25th percentile	75th percentile	P-value^*
Sex
Women	18	15	20	ns
Men	18	16	20
Age
<55	18	15	20	ns
55–65	18	15	20
>65	18	16	20
Smoking status
Ex	19	16	20	ns
Current	18	15	20
Education
Primary	18	16	20	ns
Secondary	18	16	20
University	17.5	15	19
Civil status
Married/cohabitating	18	16	20	ns
Divorced/single	18	15	20
GOLD stage
II	18	16	21	ns
III	18	16	20
IV	18	14	19
Chronic cough
No	19	16	20	029
Yes	18	15	19.5
Sputum
No	18.5	16	20	ns
Yes	18	15	20
Dyspnoea
0	18	15	21	ns
1	18	16	20
2	18	16	20
3	18	17	19
4	18	13	19
PaO2
≥8 kPa	18	16	20
<8 kPa	17.5	12	19
Number of exacerbations last year
0	18	16	20	ns
1 or 2	18	15	20
3 +	18	14	19

ns, P ≥ 0.05

Table 3

The multivariate relationship between the number of telephone calls answered, demographics and disease characteristics

	Odds ratio	95% confidence interval	P-value^*
Model 1. Demographic variables, mutually adjusted
Sex
Women	reference		ns
Men	1.06	(0.88, 1.28)
Age
<55	reference		ns
55–65	1.11	(0.82, 1.48)
>65	1.18	(0.87, 1.59)
Smoking status
Ex	reference		ns
Current	0.92	(0.76, 1.10)
Education
Primary	reference		ns
Secondary	0.97	(0.77, 1.20)
University	0.80	(0.57, 1.11)
Civil status
Married/cohabitating	reference		ns
Divorced/single	0.92	(0.75, 1.12)
Month number	0.99	(0.98, 0.99)	<0.001
Model 2. Disease characteristics, mutually adjusted
GOLD stage
II	reference		ns
III	1.02	(0.83, 1.25)
IV	0.98	(0.70, 1.37)
Chronic cough
No	reference		ns
Yes	0.85	(0.69, 1.05)
Sputum
No	reference		ns
Yes	1.05	(0.85, 1.29)
Dyspnoea grade II
No	reference		ns
Yes	0.97	(0.76, 1.24)
PaO₂
≥8.0 kPa	reference		ns
<8.0 kPa	0.76	(0.58, 1.00)
Number of exacerbations last year
0	reference		ns
1 or 2	1.06	(0.86, 1.31)
3 +	0.81	(0.56, 1.16)
Month number	0.98	(0.98, 0.99)	<0.001

ns, P ≥ 0.05

Discussion

During the two-year telephone follow up study of 381 patients with COPD, the monthly response rates were 48–81%. In multivariate analyses neither clinical nor demographic features were associated with non-response, but there were fewer answered calls during summer. To our knowledge, this is the first study to report response rates and evaluate them in the light of variables that might bias results in a telephone survey of COPD patients.

The monthly response rates were similar to response rates in telephone surveys both in Norway and other countries.^15,16 The lowest response rate was obtained in July during the second year of follow-up. In both years of follow up the response rate was lower in June and July. In the multivariate model, response rate was only dependent on the month in which the calls were placed. Non-response during summer can be explained by vacations and outdoor activities. There was a gradual drop in response rate from the first to the second year. The reduction in response rate over time could be due to lack of updated contact information.¹⁴ However, telephone numbers and addresses were updated regularly using Internet based phone directories, which are publicly available in Norway. Furthermore, this telephone survey was part of a larger cohort study of COPD patients that included regular half-yearly visits to the study centre, where the patients themselves updated contact information. Thus, the gradual drop in response rate suggests that COPD patients participating in telephone follow-up surveys tire over time. The slight increase in non-response throughout the study was neither associated with demographical nor clinical variables, which does not suggest that non-response bias increases as response rate drops. However, lower response rates can lead to decreased statistical power, and this should be kept in mind when telephone surveys of COPD patients are planned. Courses of action that might improve response rate include monetary incentives, pre-notification letters, closer follow up contact, and mentioning personal obligation to participate.¹⁷

We chose the 10th percentile as a cut off between frequent and infrequent responders. Due to the skewed distribution of the response (Figure 2), using percentiles is a useful way to describe the distribution. The median and interquartile ranges of the distribution for the subgroups of the variables were fairly similar, see Table 2. We wanted to investigate whether the infrequent responders would introduce bias, so we chose the 10th percentile as a cut off since this is often used as the lowest percentile when survey response is described.¹⁸ The finding that current smokers are inclined to respond less in unadjusted analyses is in line with findings from postal surveys^10,12,17,18 where smokers have responded less or not at all. In previous telephone surveys low response rates have been reported both as independent of^19,20 and associated with smoking.²¹ However, after adjustments for other variables, the response rate among smokers and ex-smokers did not differ significantly in our study.

In bivariate analyses subjects with chronic cough responded to fewer calls than subjects without chronic cough, which is not surprising. Subjects with chronic cough often have impaired health-related quality of life partly due to interruption of conversations or telephone calls²², and might avoid talking on the telephone.

In the multivariate analyses however, neither demographic nor disease characteristics were associated with lower response to telephone calls. The finding that response rates did not introduce bias in our study is in line with previous studies of responders and non-responders in telephone surveys^23,24, but not all.²⁵ Davern et al. concluded that no significant differences were observed in key health variables among responders and non-responders in three telephone surveys of non-institutionalized adults in Minnesota and Oklahoma.²³ Lee et al. found that response rate did not substantially affect the representativeness of data when analysing data from the California Health Interview Survey²⁴ In another American survey investigating non-response, the Behavioural Risk Factor Surveillance System (BRFSS), lower telephone response rates were related to younger individuals (18–34 years), women and racial/ethnic minorities.²⁵ Our study population was ethnically homogeneous and did not include subjects younger than 40 years of age, so we cannot evaluate these variables in relation to non-response. The American telephone surveys described above were conducted in general population samples, whereas our study was conducted in a patient cohort. Interest in the study matter positively influences response rate.²⁶ For instance, general population surveys addressing cancer have obtained higher response rate from first- or second degree-relatives of people diagnosed with cancer. All our subjects had COPD, were willing to come half yearly to the study centre and undergo testing. Their willingness to participate in order to learn more about their disease might have attenuated the effect that variables like gender and age usually have on non-response. Our study adds to the current body of knowledge by confirming that there is little risk of response bias in telephone surveys of COPD patients.

The present study had several strengths. The survey comprised all clinically relevant COPD stages, and the patients were recruited from both outpatient and general population sources. Finally, the study lasted for 24 months, enabling the descriptions of trends. However, the limitations of the study should be acknowledged. The participants were seen every six months apart from the monthly calls. How these half-yearly visits may have influenced the responses to the telephone calls is unclear. Hence the generalizability of the study results may be limited to COPD follow-up studies in which the participants are also seen at a regular basis.

In conclusion, responders and less frequent responders had similar characteristics in our telephone follow-up survey of COPD patients. This suggest that telephone surveys of COPD patients can give representative estimates for the COPD population at large, and that non-response does not necessarily introduce bias. However, response rates influence statistical power. Thus the gradual drop in response rate observed throughout our study implies that non-response and measures to counteract it should be considered when such surveys are planned.

References

Barr

, Celli

, Mannino

DM.

Comorbidities, patient knowledge, and disease management in a national sample of patients with COPD. Am J Med 2009; 122: 348–55

Bischoff

, Boer

, Molema

Validity of an automated telephonic system to assess COPD exacerbation rates. Eur Respir J 2012; 39: 1090–6

Chittleborough

, Burke

, Taylor

AW.

Medicare-related service use and costs among people with diagnosed and undiagnosed diabetes and respiratory conditions. Aust Health Rev 2009; 33: 107–16

Eisner

, Balmes

, Katz

, Trupin

, Yelin

, Blanc

PD.

Lifetime environmental tobacco smoke exposure and the risk of chronic obstructive pulmonary disease. Environ Health 2005; 4: 7

Mair

, Goldstein

, May

Patient and provider perspectives on home telecare: preliminary results from a randomized controlled trial. J Telemed Telecare 2005; 11 (Suppl. 1): 95–7

Miravitlles

, Anzueto

, Legnani

, Forstmeier

, Fargel

Patient's perception of exacerbations of COPD - the PERCEIVE study. Respir Med 2007; 101: 453–60

Sridhar

, Taylor

, Dawson

, Roberts

, Partridge

MR.

A nurse led intermediate care package in patients who have been hospitalised with an acute exacerbation of chronic obstructive pulmonary disease. Thorax 2008; 63: 194–200

Bakke

, Gulsvik

, Lilleng

, Overå

, Hanoa

, Eide

GE.

Postal survey on airborne occupational exposure and respiratory disorders in Norway: causes and consequences of non-response. J Epidemiol Community Health 1990; 44: 316–20

Eagan

, Eide

, Gulsvik

, Bakke

PS.

Nonresponse in a community cohort study: predictors and consequences for exposure-disease associations. J Clin Epidemiol 2002; 55: 775–81

10.

Hardie

, Bakke

, Morkve

Non-response bias in a postal questionnaire survey on respiratory health in the old and very old. Scand J Public Health 2003; 31: 411–7

11.

Sørheim

, Johannessen

, Grydeland

, Omenaas

, Gulsvik

, Bakke

PS.

Case-control studies on risk factors for chronic obstructive pulmonary disease: how does the sampling of the cases and controls affect the results?

Clin Respir J 2010; 4: 89–96

12.

Rönmark

, Ekerljung

, Lötvall

, Torén

, Rönmark

, Lundbäck

Large scale questionnaire survey on respiratory health in Sweden: effects of late- and non-response. Respir Med 2009; 103: 1807–15

13.

Johannessen

, Lehmann

, Omenaas

, Eide

, Bakke

, Gulsvik

Post-bronchodilator spirometry reference values in adults and implications for disease management. Am J Respir Crit Care Med 2006; 173: 1316–25

14.

Eagan

, Ueland

, Wagner

PD.

Systemic inflammatory markers in COPD: results from the Bergen COPD Cohort Study. Eur Respir J 2010; 35: 540–8

15.

Brøgger

, Bakke

, Eide

, Gulsvik

Comparison of telephone and postal survey modes on respiratory symptoms and risk factors. Am J Epidemiol 2002; 155: 572–6

16.

Groves

RM.

Nonresponse rates and nonresponse bias in household surveys. Public Opinion Quarterly 2006; 70: 646–75

17.

Bakke

PS.

Non-response in epidemiological studies - how to cope with it?

Respir Med 2010; 104: 323–4

18.

Roth

, BeVier

CA.

Response rates in HRM/OB survey research: norms and correlates, 1990-1994. J Manage 1998; 24: 97–117

19.

Verlato

, Melotti

, Olivieri

Asthmatics and ex-smokers respond early, heavy smokers respond late to mailed surveys in Italy. Respir Med 2010; 104: 172–9

20.

Kristal

, White

, Davis

JR.

Effects of enhanced calling efforts on response rates, estimates of health behavior, and costs in a telephone health survey using random-digit dialing. Public Health Rep 1993; 108: 372–9

21.

Voigt

, Koepsell

, Daling

JR.

Characteristics of telephone survey respondents according to willingness to participate. Am J Epidemiol 2003; 157: 66–73

22.

Criqui

, Barrett-Connor

, Austin

Differences between respondents and non-respondents in a population-based cardiovascular disease study. Am J Epidemiol 1978; 108: 367–72

23.

Davern

, McAlpine

, Beebe

, Ziegenfuss

, Rockwood

, Call

KT.

Are lower response rates hazardous to your health survey? An analysis of three state telephone health surveys. Health Serv Res 2010; 45: 1324–44

24.

Lee

, Brown

, Grant

, Belin

, Brick

JM.

Exploring nonresponse bias in a health survey using neighborhood characteristics. Am J Public Health 2009; 99: 1811–7

25.

Schneider

, Clark

, Rakowski

, Lapane

KL.

Evaluating the impact of non-response bias in the Behavioral Risk Factor Surveillance System (BRFSS). J Epidemiol Community Health 2010; 66: 290–5

26.

Massey

, O'Connor

, Krotki

Response Rates in Random Digit Dialing (RDD) Telephone Surveys. See http://ideas.repec.Org/p/mpr/mprres/4309.html (last checked 29 September 2012)