Pre-contact by telephone increases response rates to postal questionnaires in a population of stroke patients: an open ended randomized controlled trial

Certain kind of patient data relevant to health care researchers can only be harvested from the patients themselves, e.g. health related quality of life or treatment satisfaction. While non-subjective patient data, e.g. number of visits with a GP or hours of home care per week received, may be possible to obtain from other sources in theory, postal or electronic survey questionnaires often stand out as the cost-effective alternative ([1], p. 2). Once a survey questionnaire design is chosen, researchers should – subject to their available resources – undertake to maximize the response rate (RR). Low RR’s obviously decrease the statistical power of the final data set, increasing type II errors. Worse yet; if propensity for (non-)responding correlates with any of the collected variables, subsequent analysis will usually be biased, which increases the risk of Type I errors. This problem is particularly acute if an outcome variable is causally linked to response propensity. There is substantial evidence that responders and non-responders differs with respect to a range of characteristics; see e.g. [2] for an introduction and further references.

NORSPOT (Norwegian Stroke – Paths of Treatment) is an established cohort consisting of all patients admitted to the Stroke Unit (SU) of Akerhus University Hospital (Ahus) during the period February 15th 2012 to March 15th 2013. Ahus is situated in the Oslo greater metropolitan area, and is Norway’s largest acute hospital with a catchment area counting circa 500.000 inhabitants (≃10 % of Norway’s population). The project’s charter is to map out stroke patients’ treatment paths through the health services, and to estimate current epidemiological data, with an aim to inform policy makers about future needs within the various parts of the Norwegian health care system. To this end, in addition to data collected during hospitalization, NORSPOT seeks to elicit information about the patients’ levels of, contentment with, and benefit from, consumption of health services over the year following hospitalization. Postal survey questionnaires were therefore sent to the patients’ home address at 3 and 12 months after the time of discharge from the SU. The data set contains 1144 unique patients with a stroke diagnosis. The project recorded data on circa 1900 admissions to the SU. There were several patients with multiple stays, and about two thirds of all patients admitted to the SU received a final stroke diagnosis. Here, the stroke diagnoses are classified into intra-cranial haemorrhages (ICD10 61.X), cerebral infarctions (ICD10 63.X) and transient ischemic attacks (ICD10 45.X excluding 45.4). Patient’s with other diagnoses were not included in the follow-up questionnaire study. During the 3 month survey questionnaire phase, we observed that the response rates were converging towards the lower end of the acceptable range (≃50 %), and we were worried that further loss-to-follow-up during the 12 month questionnaire phase could weaken our study.

We therefore searched the literature on increasing response rates to postal questionnaires to see if there were any easy and efficient steps which could be implemented to maintain – or even improve – the state of affairs.

A Cochrane review from 2010 [1] identified and analysed 481 papers investigating different strategies to increase response rates to postal questionnaire surveys. We noted that amongst the diverse approaches shown to have a positive effect on response rates, many were already implemented (e.g. handwritten address, university hospital envelope), while other approaches were either considered unethical according to local ethical considerations (pecuniary incentives) or incompatible with our study (short questionnaire, non-sensitive questions).

Of the 45 papers reporting on pre-contact vs. no pre-contact on final response rates, we found very few studies targeting a population similar to our population of elderly patients. The studies were rated from A to C according to Cochrane’s standard quality assessment grading scheme ([1], p. 3). Only 5 studies were A-rated. Furthermore, the review contained information about the various studies’ target groups, and the nature of the surveys. Only 15 of the 45 papers included in [1] had a health perspective; only 8 of these targeted patients. Of the 15 health related studies, there were 6 investigating pre-contact by phone, and 9 investigating pre-contact by mail. Restricting ourselves to A- or B-rated patient-targeted pre-notification by telephone, there were one B-rated study [3] which found found a positive effect (OR = 2.58 [1.50,4.44]) of pre-contact by telephone on response rates, and one A-rated study [4] which found a positive effect (OR = 1.27 [1.00,1.63]). However, the target populations were young and mid-life women; rather dissimilar to our senescent stroke patients.

From our health services researcher’s perspective we concluded that even though the pooled data from [1] yielded a statistically significant and positive effect of pre-contact, we were unsure if it would be efficient with a fairly comprehensive questionnaire eliciting sensitive information in a population like ours. We therefore decided to undertake a randomized controlled trial (RCT), with the aim of establishing whether pre-contact by telephone increases response rates to postal questionnaires in a frail and senescent patient population.

The RCT was prospectively registered with Current Controlled Trials [5] on May 25th 2013 with International Standard RCT Number http://​www.​isrctn.​com/​ISRCTN31304930. The study was conducted between June 1st 2013 and April 1st 2014. The RCT was designed as an interventional open randomized controlled trial with two study branches – one intervention group (IG) and one control group (CG) – and main outcome was performed with a two-sided test for equality of proportion (significance level of α=0.05), analysed as per intention-to-treat (ITT). Main outcome was response rate after 45 days; secondary outcome was response rate after 365 days.

All patients discharged from the SU with a stroke diagnosis between May 15th 2012 and April 1st 2013, who:

With regard to items 1. and 3. recall that the RCT was carried out in conjunction with the second round of survey questionnaires mailed to this population. We had already defined a protocol for pre-contacting the primary caregivers of the most frail patients. As these most frail patients were also to be pre-contacted prior to mailing the 12 month questionnaire, they were not eligible for inclusion into the RCT. Item 2. was included in the protocol because we expected that the effect of the intervention would be stronger amongst those patients who were known non-responders at the 3 month follow-up (This inclusion criterion has also been employed in a B-rated study by Ogborn et. al. [6], in which they reported an OR of 1.86 in favour of pre-contact in a non-responder population).

All patients in the study received an envelope with hand-written address in the name of the patient, and, where available, also the name of the spouse. The envelope contained two questionnaires and a cover letter explaining the purpose of the study, ensuring confidentiality, and equipped with contact information for the second author (TBS) in case the recipient had any queries. The cover letter was hand-signed by TBS. The two questionnaires were made out for the patient and for a care giver; such as a spouse, a child, or another close relative or friend. The questionnaires had 16 (patient) and 8 (care giver) pages, and contained questions about history of contacts with various parts of the health care system, self reported general and mental health (modified Rankin scale, EQ-5D, HADS, Barthel ADL), knowledge about stroke, satisfaction with municipal services, and help received from/provided by care-givers. The questionnaires were estimated to take 30–60 minutes to complete. The shipment also contained two stamped and addressed return envelopes; one for the patient’s questionnaire and one for the care giver’s questionnaire. Prior to mailing the questionnaire, the status of the patient was verified with the electronic patient journal system of the hospital; i.e. whether the patient was still alive, had had a change of address or were currently admitted to hospital.

The above describes the standard protocol for the NORSPOT project.

Those who were randomized into the IG were pre-contacted by TBS according to the following protocol:

Note that the above protocol for pre-contact was already de facto established, but then aimed primarily at care givers; see item 3. in the Inclusion -section.

Each participant was assigned an internal identification number. A list of the participants identification numbers were sent to a statistician external to the project (see Acknowledgements) who generated the CG and the IG by random allocation. This was performed in two rounds: first in June 2013, and consecutively a second batch of non-responders from the three-month round were included and randomized to the CG and IG in September 2013.

The main analysis of the effect of the pre-contact intervention on response rates is performed according to the principle of intention to treat for both the primary and the outcomes. This means that the response rates calculated for the IG include those who could not be reached by telephone and patients who refused to have a questionnaire sent to them when contacted. This means that also subjects whom we tried, but were not able to, contact are recorded as in the intervention group in the main analysis. However, RCT-participants who were discovered to be either dead, terminal or diagnosed with dementia were excluded at the time of the status check which was routinely performed for all NORSPOT participants.

The primary outcome is defined as response within 45 days from the day when the questionnaire was mailed; the secondary outcome is defined as response within 365 days. Experience from the 3 month questionnaires showed that no participants had responded later than 3 months after receiving the questionnaires. The secondary outcome would therefore very likely be the absolute response rate, and would reveal it if a speed-up effect masqueraded as a true increase in final response rates. Also, for some types of surveys, the timing of the information is paramount, and old information is worthless. Therefore both outcomes can be relevant in different situations.

Return date was defined as the date written by the respondent on the questionnaire. This definition was chosen because the company handling the return of the questionnaires delivered them in batches of 10 or more at a time. This meant that some of the questionnaires would be waiting for some days prior to arriving at the hospital. We therefore decided that the date-of-completion was the most uniform measure of return.

Odds ratios and two-sided significance tests (Fisher’s exact with mid-p correction and χ²-tests) were computed with the oddsratio-function of the epitools-package [7] within the statistical environment R [8]. We report p-values w.r.t. all three tests.

Basic descriptive statistics (age, sex-distribution, type of stroke and the ratio of participants living with a spouse) for the participants were computed for the sample as a whole, and for the IG and the CG separately: The latter values were also tested for significant differences with a standard two-sided t-test. The variable living with a spouse was included because the main protocol for mailing questionnaires involved printing the name of the spouse in addition to the name of the patient wherever a registered spouse existed on the same address. This practise originated in a hope that it would be more likely that the questionnaires were opened and returned when two individuals were named on the envelope.

A binary logistic model was also fitted to the combined data set, with responding as the dependent variable, and with predictors age, sex, type of stroke, a dummy for living with a spouse, and a dummy for being in the pre-contact group (i.e. by ITT).

Finally, a simple Cost Utility Analysis (CUA) reporting the cost per additional reply associated with this strategy for increasing the response rates is also provided.

The NOR-SPOT project has been submitted to a regional ethics committee for medical research – Regional Committee for Medical & Health Research Ethics South East Norway, Section B – and exempted from evaluation because the study hypotheses were not directly related to health and illness. Under Norwegian law the project should then be approved by the hospital’s internal Privacy Ombudsman. The RCT was approved prior to initation of recruitment, with Ref.No. 11.076. Any queries should be directed to the Privacy Ombudsman at Akershus University Hospital.