Archives of Physical Medicine and Rehabilitation
Volume 90, Issue 9 , Pages 1499-1505, September 2009

Predictors of Nonresponse in a Questionnaire-Based Outcome Study of Vocational Rehabilitation Patients

  • Cyrille Burrus, MD

      Affiliations

    • Clinique romande de réadaptation suvaCare, Sion, Switzerland
    • ,
  • Pierluigi Ballabeni, PhD

      Affiliations

    • Clinique romande de réadaptation suvaCare, Sion, Switzerland
    • Institute of Social and Preventive Medicine, Centre Hospitalier Universitaire Vaudois and University of Lausanne, Lausanne, Switzerland
    • Corresponding Author InformationReprint requests to Pierluigi Ballabeni, PhD, Clinique romande de réadaptation, Case postale 352, Avenue Grand-Champsec 90, CH-1951 Sion, Switzerland
    • ,
  • Olivier Deriaz, MD, PhD

      Affiliations

    • Clinique romande de réadaptation suvaCare, Sion, Switzerland
    • ,
  • Charles Gobelet, MD

      Affiliations

    • Clinique romande de réadaptation suvaCare, Sion, Switzerland
    • ,
  • François Luthi, MD

      Affiliations

    • Clinique romande de réadaptation suvaCare, Sion, Switzerland

Article Outline

Abstract 

Burrus C, Ballabeni P, Deriaz O, Gobelet C, Luthi F. Predictors of nonresponse in a questionnaire-based outcome study of vocational rehabilitation patients.

Objective

To identify predictors of nonresponse to a self-report study of patients with orthopedic trauma hospitalized for vocational rehabilitation between November 15, 2003, and December 31, 2005. The role of biopsychosocial complexity, assessed using the INTERMED, was of particular interest.

Design

Cohort study. Questionnaires with quality of life, sociodemographic, and job-related questions were given to patients at hospitalization and 1 year after discharge. Sociodemographic data, biopsychosocial complexity, and presence of comorbidity were available at hospitalization (baseline) for all eligible patients. Logistic regression models were used to test a number of baseline variables as potential predictors of nonresponse to the questionnaires at each of the 2 time points.

Setting

Rehabilitation clinic.

Participants

Patients (N=990) hospitalized for vocational rehabilitation over a period of 2 years.

Interventions

Not applicable.

Main Outcome Measure

Nonresponse to the questionnaires was the binary dependent variable.

Results

Patients with high biopsychosocial complexity, foreign native language, or low educational level were less likely to respond at both time points. Younger patients were less likely to respond at 1 year. Those living in a stable partnership were less likely than singles to respond at hospitalization. Sex, psychiatric, and somatic comorbidity and alcoholism were never associated with nonresponse.

Conclusions

We stress the importance of assessing biopsychosocial complexity to predict nonresponse. Furthermore, the factors we found to be predictive of nonresponse are also known to influence treatment outcome and vocational rehabilitation. Therefore, it is important to increase the response rate of the groups of concern in order to reduce selection bias in epidemiologic investigations.

Key Words: Questionnaires, Rehabilitation, Selection bias

List of Abbreviations: OR, odds ratio, VIF, variance inflation factor

 

EPIDEMIOLOGIC INVESTIGATIONS based on self-report questionnaires are usually characterized by the lack of response of some of the recruited subjects. If the probability of responding is associated with some exposure variable or the outcome under study, the study results will be biased. A high response proportion alone is not sufficient to guarantee unbiased results, because bias depends mainly on differences between responders and nonresponders.1, 2 Because nonresponder characteristics are usually not known or barely known,3, 4 the conclusions drawn from studies may sometimes be incorrect.

Nonresponders are scarcely described in the rehabilitation literature. When precisions are given, socioeconomic factors such as education,5, 6, 7 income,6, 7 sex,5, 7 language,8, 9 or age9 appear to distinguish them from responders. We found no data about nonresponder biopsychosocial complexity in the rehabilitation literature. In other specialties, previous research trying to define nonresponders' characteristics mostly examined sociodemographic data, retrospectively retrieved in population registries, or study modality.1, 4, 10, 11 Factors like age, sex, marital status, education level, ethnicity, contact modes, use of incentives, psychologic factors, or mother tongue have been found to be associated with nonresponse. Results differ among investigations, pathologies, and study characteristics. For instance, older age, low self-reported pain, and few contacts with health services were factors of nonparticipation in a study of patients with rheumatoid arthritis12; male, young, single or divorced, and less educated people were less likely to participate in a questionnaire study of neck and chest pain13; in a prospective osteoporosis study, nonresponders were more likely to be women, to be older, and to have higher blood pressure.14

The primary aim of the present investigation was to analyze a number of potential predictors of nonresponse to a self-report study of patients hospitalized for vocational rehabilitation. For this purpose, we used data from responders and nonresponders of a prospective cohort study performed at our clinic on patients hospitalized for rehabilitation after traumatic injury between 2003 and 2005. The aim of that study, called OUTCOME, was a longitudinal assessment of the social and psychologic outcome of patients; health care consumption; and their return to work in relation to diagnosis, therapy volume, biopsychosocial complexity, and other baseline data. All eligible patients were asked to fill out self-report questionnaires at hospitalization, at discharge, and 1 year after discharge. As potential predictors of nonresponse to the OUTCOME study, we used baseline variables that were assessed on all eligible patients. We were particularly interested in testing the role of INTERMED, an integrated measure of biopsychosocial complexity,15, 16, 17 as a potential predictor of nonresponse.

Until recently, biopsychosocial assessment has been conducted by using several distinct questionnaires. This is difficult in everyday practice because it is time-consuming, and patients must have sufficient education to fill out the questionnaires.18 INTERMED was developed to assess the patients' biopsychosocial complexity in order to operationalize the biopsychosocial model of health care by Engel.19, 20 INTERMED synthesizes data from 4 domains (biologic, psychologic, social, health care system), with each one assessed over time (past, present, prognosis) during a structured medical interview of 20 minutes. The interview is conducted by trained staff. The INTERMED contains 5 questions for each domain (table 1). Each question is rated on a 4-point scale from 0 to 3. A total INTERMED score ranging from 1 to 60 is calculated, whereby a higher score means a higher biopsychosocial complexity. For everyday practice with a prospective goal, a cut-off score of 21 was established, differentiating complex from simple cases.21, 22, 23, 24, 25

Table 1. INTERMED Domains and Variables
HistoryCurrent StatePrognoses
Biologic
Chronicity

Diagnostic dilemma


Severity of symptoms

Diagnostic challenge

Complications and life threat
Psychologic
Restrictions in coping

Psychiatric dysfunction


Resistance to treatment

Psychiatric symptoms

Mental health threat
Social
Restrictions in integration

Social dysfunctioning


Residential instability

Restrictions of network

Social vulnerability
Health care
Intensity of treatment

Treatment experience


Organization of care

Appropriateness of referral

Coordination

From Huyse FJ, Lyons JS, Stiefel FC, et al. “INTERMED”: a method to assess health service needs, I: development and reliability. Gen Hosp Psychiatry 1999;21:39-48.15 Reprinted with permission.

INTERMED has been compared with a variety of other validated instruments such as the Medical Outcomes Study 36-Item Short-Form Health Survey, the Hospital and Anxiety and Depression Scale, the visual analog scale, and other scales measuring social domain. It showed high interrater reliability and agreement.15, 26 Predictive validity was studied in different patient populations. Among patients with low back pain,16, 27 the INTERMED score was significantly higher in patients applying for disability compensation, showing ability to identify patients with a chronic and disabling course. In patients admitted in a medical ward, a high INTERMED score was associated with longer hospital stay, increased use of medications, nurse intervention, or specialist interventions.26 In patients with diabetes, high INTERMED scores were associated with increased health care use.28 In patients diagnosed with multiple sclerosis, a high score was associated with the number of disciplines proposed in the multidisciplinary treatment plan.29 It was also studied in patients with rheumatoid arthritis25 and patients receiving dialysis,23 where it confirmed its ability to detect complex cases.

Back to Article Outline

Methods 

Study Population 

The study sample consisted of 990 patients hospitalized between November 15, 2003, and December 31, 2005, after orthopedic trauma, eligible for the OUTCOME study. This is a prospective cohort study of patients enrolled in 2 Swiss rehabilitation clinics after traumatic injury. Clinical and sociodemographic data at hospitalization (baseline) were available for all the 990 eligible patients. A semiconducted interview was organized for every admitted patient during the first 3 days of admission to evaluate biopsychosocial complexity. Nonresponse to self-report questionnaires was analyzed for patients with orthopedic trauma from only 1 of the 2 clinics, where INTERMED is systematically used by the orthopedic rehabilitation unit. We excluded from the OUTCOME study patients with severe traumatic brain injury (Glasgow Coma Scale≤8), with spinal cord injury, with insufficient judgment capacity, or under legal custody. All other patients coming for rehabilitation after a traumatic injury were eligible. If a patient was hospitalized more than once during the study period, we considered only the first stay for this analysis.

Most of our inpatients are blue-collar workers and come to our facility after work or traffic collisions (table 2). There was no predominant main site of injury. Knee (fractures, ligament injuries, posttraumatic osteoarthritis), spine (fractures, posttraumatic low back pain), and shoulder (fractures, tendon injuries, posttraumatic adhesive capsulitis) count each for almost 20%, followed by foot/ankle (fractures, sprains, complex regional pain syndrome) and hand/wrist injury (fractures, sprains, complex regional pain syndrome). At the time of hospitalization, nearly all patients lived in the French part of Switzerland, where our clinic is located, or in nearby France (frontier workers). All patients were fluent in spoken French and had the choice of answering the OUTCOME questionnaires either in French or in German, the 2 most common national languages of Switzerland.

Table 2. Patient Characteristics Related to Accident
Responders (n=648)Nonresponders (n=342)P
Frequency%Frequency%
Circumstance of trauma
Work26741.222164.6Pearson χ2 test for contingency table: <.001
Leisure8713.4133.8
Domestic8212.63510.2
Road traffic15023.25114.9
Other629.6226.4
MedianMin-maxMedianMin-max
Time from accident to admission into rehabilitation clinic (mo)90.3–51380.4–241Mann-Whitney U test: .023
Days in the rehabilitation clinic292–128281–1052-Sided t test: .002

Abbreviations: Max, maximum; min, minimum.

Our clinic is specialized in the treatment and evaluation of patients after traumatic injuries. Patients are hospitalized when they present persistent pain and functional limitations after an accident (median=9mo after the accident). The aim of the therapeutic program is to take care of patients with a multidisciplinary approach (somatic and psychologic) in order to improve patient quality of life, functional status, and chance of returning to work. At the end of the hospitalization (median duration=29d) a program is defined in order to plan a return to the former workplace, which may sometimes be adapted to the disability. If necessary, other medical measures are planned, such as new surgery. If disabilities appear to limit return to former work, the goal of hospitalization is to determine functional limitations. In this situation, the patient can then be taken in charge by the Federal Disability Insurance, the Swiss national insurance responsible for vocational rehabilitation.

Data 

Within 3 days posthospitalization, all eligible patients were given a written description of the OUTCOME study and asked whether they wanted to participate. In the present investigation, we considered nonresponders all eligible patients who did not fill out the OUTCOME questionnaire at hospitalization or 1 year after discharge, whether or not they took part in the OUTCOME study. Nonparticipants were treated as nonresponders. Those who agreed to participate in the OUTCOME protocol signed an informed consent form. All data were treated in conformity with the Swiss federal law on data protection. The OUTCOME protocol was approved by the ethical committee of the local medical association.

We tested age, sex, marital status, educational level, native language, biopsychosocial complexity, psychiatric comorbidity, alcoholism, and somatic comorbidity as potential predictors of nonresponse. Comorbidities and biopsychosocial complexity were assessed by a trained physician, based on health conditions during the entry interview. Psychiatric diagnoses were confirmed by a specialist. Comorbidities were assessed following the International Statistical Classification of Diseases and Related Health Problems, 10th revision. For the evaluation, comorbidities were considered present or not. Treating personnel collected these data within 3 days after hospitalization.

As a measure of educational level, we distinguished between patients who only went to mandatory school and patients who had some further education beyond mandatory school. The duration of compulsory school is 9 years in Switzerland but can be slightly longer or shorter for patients who grew up abroad. We did not ask patients for more precise information.

Details on the semistructured interview can be obtained at http://www.intermedfoundation.org/manual/im5eng.pdf. The scoring system allows a maximum score of 15 for each of the 4 domains, which can be summed to a total score ranging from 0 to 60, reflecting the complexity of the case.

Statistical Analysis 

We used multiple logistic regressions to estimate the association between the dependent binary variable called response and the potential predictors listed. The use of multiple regressions allowed assessing the effects of each predictor while adjusting for the other predictors tested. Age was analyzed as a continuous variable expressed in years, while all other predictors were dichotomous (see reference categories in table 3). In particular, we dichotomized the total INTERMED score to obtain simple patients (total score<21) versus complex ones (total score≥21), because this is the way the INTERMED is used by practitioners.16, 27 We coded response as 1 if a subject did not respond and 0 if a subject responded. Therefore, a higher OR implies a higher probability of not responding to the OUTCOME questionnaire. We performed the analyses separately for the questionnaires collected at hospitalization and 1 year after discharge from the clinic. Wald tests were used to calculate P values in the logistic regressions.

Table 3. Frequency of the Predictor Categories, Nonresponse Proportions, and Final Logistic Regression Models for Nonresponse (N=990 for all variables)
VariableCategoryFrequency (%)Nonresponse at HospitalizationNonresponse 1 Year After Discharge
Nonresponder Frequency (%)OR95% CIPNonresponder Frequency (%)OR95% CIP
Age 0.98⁎0.97–1.00.012
SexMen (ref)865(87.4)307(35) 557(64)
Women125(12.6)35(28) 75(60)
Marital statusMarried or domestic partner (ref)651(65.8)272(42)1 431(66)
Single339(34.2)70(21)0.530.37–0.76.001201(59)
EducationCompulsory school (ref)517(52.2)271(52)1 391(76)1
Further education473(47.8)71(15)0.340.24–0.49<.001241(51)0.560.41–0.76<.001
Native languageFrench or German (ref)458(46.3)36(8)1 209(46)1
Other532(53.7)306(58)9.786.55–14.50<.001423(80)3.402.50–4.62<.001
Biopsychosocial complexitySimple (ref)473(47.8)120(25)1 278(57)1
Complex517(52.2)222(43)1.991.44–2.76<.001364(70)1.501.14–2.00.004
Psychiatric comorbidityAbsent (ref)737(74.4)229(31) 457(62)
Present253(25.6)113(45) 175(69)
AlcoholismAbsent (ref)917(92.6)319(35) 589(64)
Present73(7.3)23(32) 43(59)
Somatic ComorbidityAbsent (ref)641(64.7)215(34) 410(64)
Present349(35.3)127(36) 222(64)

NOTE. A higher OR means a higher probability of being a nonrespondent. Only those variables included in the final model have ORs and CIs.

Abbreviations: CI, confidence interval; ref, reference category.

⁎OR for increases of 1 year.

We built statistical models as follows.30 First, we tested each predictor separately in a simple logistic regression. Second, we tested all predictors that reached a significance level 0.2 or less together in a first multiple logistic model (the full model). Third, in a backward stepwise selection procedure, we dropped predictors that were nonsignificant at the .05 level in the preceding model, starting from that with the highest P value, until only statistically significant predictors remained. Each time a variable was dropped, we compared the new model with the full model via likelihood ratio tests to make sure that the model fit had not decreased significantly. We assessed the final models for goodness of fit via the Hosmer-Lemeshow statistic30 and lowess graph of predicted probabilities against observed values.31 To check for multicollinearity, we calculated the VIFs for the independent variables of the full and the final models. In logistic regression, a VIF above 2.5 may indicate the presence of a multicollinearity problem.

To internally validate the models, we assessed their predictive discrimination performance by estimating unbiased, bootstrap-corrected c statistics.32, 33 In our analyses, the c statistic measures the ability of a model to discriminate subjects with high probability of responding from those with low probability and is equivalent to the area under the receiver operating characteristic curve. The c statistic ranges from 0.5 (no predictive discrimination) to 1 (perfect discrimination); values below 0.7 indicate a poor discrimination ability, values 0.7 to 0.8 a fair discrimination, and values above 0.8 a good to excellent discrimination. We ran 200 bootstrap resamples (N=990 each) to calculate unbiased c statistics for the final and the full models.33 We preferred bootstrapping to other internal validation methods because the former has shown more accurate and less biased estimations.33

All analyses were performed with the statistical software Stata 9.2.a

Back to Article Outline

Results 

Nine hundred ninety patients were eligible for the OUTCOME study and were therefore included in the present investigation. Their mean age ± SD was 42±10.7 years (range, 17–64y). Other characteristics are shown in table 2. The more frequent categories in each of the dichotomous predictors were as follows: men (87%), married or domestic partner (66%), compulsory school (52%), native language other than French or German (54%), complex biopsychosocial case (52%), lack of psychiatric comorbidity (74%), lack of alcoholism (93%), and lack of somatic comorbidity (65%) (see table 3). French was the native language of 439 patients (44% of the total sample) and German that of 19 (2%). The most frequent of the other languages were Portuguese (n=185; 19%), Italian (n=94; 9%), Albanese (n=75; 8%), Serbian (n=40; 4%), and Spanish (n=37; 4%).

Nonresponse at Hospitalization 

Six hundred forty-eight (65%) of the eligible patients accepted to participate to the study and thus responded to the questionnaires at hospitalization. Responders and nonresponders had the same mean age (42.4 and 42.9y, respectively). The numbers and percent of responders in each predictor category are shown in table 3.

As a result of the univariate regressions, sex, marital status, education, native language, psychiatric comorbidity, and biopsychosocial complexity were entered in the full multiple regression model. The stepwise procedure led to the drop of sex and psychiatric comorbidity. Thus, marital status, education, native language, and biopsychosocial complexity were retained in the final model. Nonresponse was more likely in patients living in a stable partnership (OR=.53 for single living), having lower education (OR=.34 for education beyond compulsory school level), speaking a language different than French or German (OR=9.78), or being a complex biopsychosocial case (OR=1.99) (see table 3).

The final model did not fit the data significantly worse than the full model (likelihood ratio test: χ2=2.88, df=2; P=.237). The Hosmer-Lemeshow statistic showed no evidence that the final model failed at predicting the data (χ2=5.94, df=8; P=.654). Finally, the discrimination performance of the final model was rather good, with a bootstrap-corrected c equal to .837, while the full model had c equal to .841. Multicollinearity was not an issue in either model (all VIFs<1.3).

Nonresponse After 1 Year 

Three hundred fifty-eight patients (36% of the initial 990) responded to the questionnaire 1 year after discharge from the clinic. The mean age ± SD of the responders was 44.4±10.9 years, and that of nonresponders, 41.9±10.6 years. The numbers and percentages for the dichotomous predictors can be seen in table 3.

After the univariate regressions, age, marital status, education, native language, psychiatric comorbidity, and biopsychosocial complexity were used to build the full multiple regression model. Marital status and psychiatric comorbidity were then dropped, leaving a final model including age, education, native language, and biopsychosocial complexity (see table 3). Thus, nonresponse was more likely in younger patients (OR=.98 for each increase of 1 year), and again in those with lower educational status (OR=.56 for education beyond compulsory school level), language different than French or German (OR=3.40), and complex biopsychosocial case (OR=1.50).

Again, the fit of the final model was not significantly worse than that of the full model (likelihood ratio test: χ2=1.22, df=2; P=.543). The Hosmer-Lemeshow statistic showed no evidence that the final model failed at predicting the data (χ2=4.68, df=8; P=.791). The bootstrap-corrected c of the final model was .727, while that of the full model was .729. Again, no multicollinearity was present (all VIFs<1.4 in both models).

Back to Article Outline

Discussion 

Our results suggest that biopsychosocial complexity, language, educational status, and, to a lesser extent, marital status and age may influence nonresponse in a questionnaire-based investigation of vocational rehabilitation study. Sex, psychiatric comorbidity, alcoholism, and somatic comorbidity had no significant effect in this investigation. The effects of biopsychosocial complexity, language, and education were stronger at hospitalization but were still important over 1 year after their assessment. All the potential predictors we tested were assessed prospectively, and their values were available for all responders and nonresponders.

Our results strongly suggest that, in a rehabilitation program, independent from other factors, patients with high biopsychosocial complexity are less likely than simple patients to respond to questionnaire-based studies, with the potential risk of biasing the study results. Previous studies showed that patients classified as complex by the INTERMED were at higher risk for poor outcome or increased health care use among patients with low back pain or in rheumatology.27, 34 Complex patients with low back pain had high symptom persistence, high prevalence of psychosocial morbidities, and low rate of return to work,27 and complex patients with rheumatic arthritis had more hospitalizations or medical consultations and lower quality of life.34 Thus, INTERMED-complex subjects being less likely to respond to surveys and having more negative outcomes means that self-report studies of the rehabilitation outcome are likely to produce overoptimistic results. We therefore stress the importance of assessing, when feasible, the biopsychosocial complexity of potential study participants, and of making an effort to facilitate the participation of complex cases. The INTERMED seems to us to be a unique and simple tool to assess biopsychosocial complexity as a whole, summarizing in 1 single score biological, psychologic, social, and medical factors.

We also found a native language different from that of the OUTCOME questionnaires (ie, French or German) to be an independent predictor of nonresponse, both at hospitalization and 1 year after discharge. Such a language effect is not new but, unlike in several other investigations,4, 35, 36, 37 we were able to include language in multiple regressions, therefore adjusting its effect by that of other predictors and vice versa. In the present investigation, the effect of language was therefore independent from biopsychosocial complexity, educational status, marital status, and age. This suggests that nonresponse was indeed a result of difficulties with the written language, although all eligible patients were judged fluent in spoken French by the nurses who interviewed them all for the INTERMED.

Language is the most powerful predictor in our study both at hospitalization (OR=9.78) and after 1 year (OR=3.4). With the present worldwide increasing number of migrants, more people are potential subjects of studies requiring answers to questionnaires written in a foreign language. Switzerland, where the present study was performed, is characterized by 4 national languages spoken in distinct geographical areas and by a large amount of immigration, with 22% of the resident population born abroad.38 In the United States, 11% of the population was born abroad, and 16% speak a language different than English at home.39 In the European Union, 23 national languages are spoken, causing linguistic vulnerability among European Union migrants as well as among those coming from outside the European Union. There is presently a poor choice of validated questionnaires translated in different languages. Because foreign speakers are usually a more vulnerable population,40 their lower questionnaire response proportion is very likely to bias an investigation, hence the importance of providing questionnaire translations to improve participation even in countries with only 1 official language. Moreover, more knowledge is needed on cultural differences in perceptions and concepts between ethnic groups.

A lower educational status was also found to be a predictor of nonresponse, confirming what was found in some previous studies.3 Our results were adjusted for biopsychosocial complexity, native language, marital status, and age, therefore excluding confounding by these variables. Patients with a lower education, defined here as having had not more than compulsory school level, may have had more difficulties and needed more time to answer questionnaires. This might be related to a poor reading ability. Reading problems may lead not only to poor socioprofessional reinsertion but also to bad therapeutic compliance.41 Extra effort could be made to encourage the participation of patients with lower education, maybe by choosing, when possible, shorter validated questionnaires.

High biopsychosocial complexity, foreign native language, and low education had an effect both at hospitalization and 1 year after discharge. On the contrary, age and marital status were associated with nonresponse in only 1 of these occasions. Younger people were less likely to respond 1 year after discharge but not at hospitalization. Young age seems to characterize nonresponders more often in the literature,9, 13, 42, 43, 44 but depending on the context, old age has also been found to be a predictor of nonresponse in questionnaires studies.12, 14, 45, 46 In our study, young age is a predictor of nonresponse after 1 year, but not during hospitalization. We have no clear explanation of this, but context it might influence results, with the first questionnaires given during hospitalization and the second sent home 1 year later. Marital status was predictive of nonresponse at hospitalization but not at 1 year, with patients living as a couple having a lower likelihood of nonresponse. This is not in agreement with previous studies, in which the nonrespondents were more likely to be living single,13, 43, 44, 47 but the setting was different. We have no explanation for this, except that patients who used to live as a couple may have more difficulties adapting to their hospital stay and my find less time to answer questionnaires.

Study Limitations 

This is the first study specifically designed to address nonresponders' characteristics in rehabilitation, and therefore needs to be confirmed by other investigations.

The variables we tested as nonresponse predictors were carefully chosen prior to the analysis among several available baseline variables. However, we cannot be sure that all relevant potential confounding variables were included in our statistical models. The absence of important confounding variables results in biased parameter estimates in logistic regression models.

Our findings can be generalized only to patients hospitalized for rehabilitation after orthopedic trauma, who have a different profile than patients hospitalized for acute care. Our patients were mainly male blue-collar workers.

Education level was assessed by asking patients what school degree they reached (compulsory, vocational qualification, high school, university). We were not able to quantify precisely the duration of education, and we did not try to make more precise education categories because the patients were raised in different countries without comparable school systems. Furthermore, we assessed speaking fluency but not reading proficiency. Evaluation of the latter would have provided useful information.

All statistical model building techniques can be criticized because they bear some arbitrary and subjective elements.48 We internally validated the performance of our statistical models, but the most definitive test of a statistical model would be external validation, requiring testing of our models on data sets from different populations.

Back to Article Outline

Conclusions 

The factors we found to be predictive of nonresponse, particularly low educational level, foreign native language, and high biopsychosocial complexity, are also known to influence treatment outcome and vocational rehabilitation.25, 27, 28, 49, 50 In order to reduce selection bias and limit the risk of misinterpreting results of questionnaire-based studies, an extra effort should be made to improve the likelihood of response by people who are biopsychosocially complex, speak foreign languages, and have low educational status, particularly in the context of vocational rehabilitation. Therefore, we stress the importance of systematically assessing biopsychosocial complexity. The INTERMED is an easy method for this purpose. Validated translations of questionnaires should be made available in countries with a multicultural population even if the potential study participants are likely to be fluent in the local official language. Questionnaires should be kept short to avoid discouraging people who are not comfortable with paperwork.

Supplier

Back to Article Outline

Acknowledgments 

We thank Viviane Dufour, Antoinette Crettenand, and Aaron Russell, PhD, for all the work related to the preparation of the OUTCOME questionnaires, data collection, and data entry. Peter Erhard, PhD, Hanspeter Gmünder, MD, and Nikola Seichert, PhD, collaborated in the design of the OUTCOME project.

Back to Article Outline

References 

  1. Stang A. Nonresponse research–an underdeveloped field in epidemiology. Eur J Epidemiol. 2003;18:929–931
  2. Vestbo J, Rasmussen FV. Baseline characteristics are not sufficient indicators of non-response bias in follow-up studies. J Epidemiol Community Health. 1992;46:617–619
  3. Jacomb PA, Jorm AF, Korten AE, Christensen H, Henderson AS. Predictors of refusal to participate: a longitudinal health survey of the elderly in Australia. BMC Public Health. 2002;2:4
  4. Johnson TP, Holbrook AL, Cho YI, Bossarte RM. Nonresponse error in injury-risk surveys. Am J Prev Med. 2006;31:427–436
  5. Butcher JL, MacKenzie EJ, Cushing B, et al. Long-term outcomes after lower extremity trauma. J Trauma. 1996;41:4–9
  6. Michaels AJ, Michaels CE, Moon CH, Zimmerman MA, Peterson C, Rodriguez JL. Psychosocial factors limit outcomes after trauma. J Trauma. 1998;44:644–648
  7. Michaels AJ, Michaels CE, Smith JS, Moon CH, Peterson C, Long WB. Outcome from injury: general health, work status, and satisfaction 12 months after trauma. J Trauma. 2000;48:841–848discussion 8-50
  8. Harris IA, Young JM, Rae H, Jalaludin BB, Solomon MJ. Predictors of general health after major trauma. J Trauma. 2008;64:969–974
  9. Ponsford J, Hill B, Karamitsios M, Bahar-Fuchs A. Factors influencing outcome after orthopedic trauma. J Trauma. 2008;64:1001–1009
  10. Groves RM. Nonresponse rates and nonresponse bias in household survey. Public Opin Q. 2006;646–675
  11. Stang A, Jöckel KH. Studies with low response proportions may be less biased than studies with high response proportions. Am J Epidemiol. 2004;159:204–210
  12. Rupp I, Triemstra M, Boshuizen HC, Jacobi CE, Dinant HJ, van den Bos GA. Selection bias due to non-response in a health survey among patients with rheumatoid arthritis. Eur J Public Health. 2002;12:131–135
  13. Fejer R, Hartvigsen J, Kyvik KO, Jordan A, Christensen HW, Hoilund-Carlsen PF. The Funen Neck and Chest Pain study: analysing non-response bias by using national vital statistic data. Eur J Epidemiol. 2006;21:171–180
  14. Heilbrun LK, Ross PD, Wasnich RD, Yano K, Vogel JM. Characteristics of respondents and nonrespondents in a prospective study of osteoporosis. J Clin Epidemiol. 1991;44:233–239
  15. Huyse FJ, Lyons JS, Stiefel FC, et al. “INTERMED”: a method to assess health service needs, I: development and reliability. Gen Hosp Psychiatry. 1999;21:39–48
  16. Stiefel FC, de Jonge P, Huyse FJ, et al. “INTERMED”: a method to assess health service needs, II: results on its validity and clinical use. Gen Hosp Psychiatry. 1999;21:49–56
  17. Stiefel FC, Huyse FJ, Sollner W, et al. Operationalizing integrated care on a clinical level: the INTERMED project. Med Clin North Am. 2006;90:713–758
  18. Ponzer S, Bergman B, Brismar B, Johansson LM. A study of patient-related characteristics and outcome after moderate injury. Injury. 1996;27:549–555
  19. Engel GL. The need for a new medical model: a challenge for bio-medicine. Science. 1977;196:129–136
  20. Engel GL. The clinical application of the biopsychosocial model. Am J Psychiatry. 1980;137:535–544
  21. de Jonge P, Bauer I, Huyse FJ, Latour CH. Medical inpatients at risk of extended hospital stay and poor discharge health status: detection with COMPRI and INTERMED. Psychosom Med. 2003;65:534–541
  22. de Jonge P, Hoogervorst EL, Huyse FJ, Polman CH. INTERMED: a measure of biopsychosocial case complexity: one year stability in multiple sclerosis patients. Gen Hosp Psychiatry. 2004;26:147–152
  23. de Jonge P, Ruinemans GM, Huyse FJ, ter Wee PM. A simple risk score predicts poor quality of life and non-survival at 1 year follow-up in dialysis patients. Nephrol Dial Transplant. 2003;18:2622–2628
  24. Di Gangi Herms AM, Pinggera GM, de Jonge P, Strasser H, Sollner W. Assessing health care needs and clinical outcome with urological case complexity: a study using INTERMED. Psychosomatics. 2003;44:196–203
  25. Koch N, Stiefel F, de Jonge P, et al. Identification of case complexity and increased health care utilization in patients with rheumatoid arthritis. Arthritis Rheum. 2001;45:216–221
  26. de Jonge P, Latour C, Huyse FJ. Interrater reliability of the INTERMED in a heterogeneous somatic population. J Psychosom Res. 2002;52:25–27
  27. Stiefel FC, de Jonge P, Huyse FJ, et al. INTERMED: an assessment and classification system for case complexity: results in patients with low back pain. Spine. 1999;24:378–384discussion 85
  28. Fischer CJ, Stiefel FC, de Jonge P, et al. Case complexity and clinical outcome in diabetes mellitus: a prospective study using the INTERMED. Diabetes Metab. 2000;26:295–302
  29. Hoogervorst EL, de Jonge P, Jelles B, et al. The INTERMED: a screening instrument to identify multiple sclerosis patients in need of multidisciplinary treatment. J Neurol Neurosurg Psychiatry. 2003;74:20–24
  30. Hosmer DW, Lemeshow S. Applied logistic regression. 2nd ed.. New York: John Wiley and Sons; 2000;
  31. Long JS, Freese J. Regression models for categorical variables using Stata. 2nd ed.. College Station: Stata Pr; 2006;
  32. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387
  33. Steyerberg EW, Harrell FE, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol. 2001;54:774–781
  34. Scerri M, de Goumoens P, Fritsch C, Van Melle G, Stiefel F, So A. The INTERMED questionnaire for predicting return to work after a multidisciplinary rehabilitation program for chronic low back pain. Joint Bone Spine. 2006;73:736–741
  35. Dickinson H, Parkinson K, McManus V, et al. Assessment of data quality in a multi-centre cross-sectional study of participation and quality of life of children with cerebral palsy. BMC Public Health. 2006;6:273
  36. Oremus M, Wolfson C. Female specialists were more likely to respond to a postal questionnaire about drug treatments for Alzheimer disease. J Clin Epidemiol. 2004;57:620–623
  37. Stang A, Moebus S, Dragano N, et al. Baseline recruitment and analyses of nonresponse of the Heinz Nixdorf Recall Study: identifiability of phone numbers as the major determinant of response. Eur J Epidemiol. 2005;20:489–496
  38. Wanner P. Recensement fédéral de la population 2000: migration et intégration. Neuchâtel: Office fédéral de la statistique; 2004;
  39. Census 2000 summary file 3 (Technical documentation). Washington (DC): US Census Bureau; 2002;
  40. Weech-Maldonado R, Elliott MN, Morales LS, Spritzer K, Marshall GN, Hays RD. Health plan effects on patient assessments of Medicaid managed care among racial/ethnic minorities. J Gen Intern Med. 2004;19:136–145
  41. de Moerloose P, Mazzolai L, Bounameaux H. Patient education and illiteracy in Switzerland. Rev Med Suisse. 2008;4:323–324
  42. Angus VC, Entwistle VA, Emslie MJ, Walker KA, Andrew JE. The requirement for prior consent to participate on survey response rates: a population-based survey in Grampian. BMC Health Services Research. 2003;3:21
  43. Lundberg I, Damström-Thakker K, Hällström T, Forsell Y. Determinants of non-participation, and the effects of non-participation on potential cause-effect relationships, in the PART study on mental disorders. Soc Psychiatry Psychiatr Epidemiol. 2005;40:475–483
  44. Sogaard AJ, Selmer R, Bjertness E, Thelle D. The Oslo Health Study: the impact of self-selection in a large, population-based survey. Int J Equity Health. 2004;3:3
  45. Barton J, Bain C, Hennekens CH, et al. Characteristics of respondents and non-respondents to a mailed questionnaire. Am J Public Health. 1980;70:823–825
  46. O'Neill TW, Marsden D, Silman AJ. Differences in the characteristics of responders and non-responders in a prevalence survey of vertebral osteoporosis (European Vertebral Osteoporosis Study Group). Osteoporos Int. 1995;5:327–334
  47. Reijneveld SA, Stronks K. The impact of response bias on estimates of health care utilization in a metropolitan area: the use of administrative data. Int J Epidemiol. 1999;28:1134–1140
  48. Rothman KJ, Greenland S, Lash TL. Modern epidemiology. 3rd ed.. Philadelphia: Lippincott, Williams and Wilkins; 2008;
  49. MacKenzie EJ, Morris JA, Jurkovich GJ, et al. Return to work following injury: the role of economic, social, and job-related factors. Am J Public Health. 1998;88:1630–1637
  50. Seland K, Cherry N, Beach J. A study of factors influencing return to work after wrist or ankle fractures. Am J Ind Med. 2006;49:197–203
  • a Version 9; StataCorp LP, 4905 Lakeway Dr, College Station, TX 77845.

 Supported by Schweizerische Unfallversicherungsanstalt (SUVA) (grant no. 100204).

 A commercial party having a direct financial interest in the results of the research supporting this article has conferred or will confer a financial benefit on 1 or more of the authors. The Clinique romande de réadaptation belongs to SUVA, the Swiss national accident insurance.

PII: S0003-9993(09)00356-6

doi:10.1016/j.apmr.2009.03.014

Archives of Physical Medicine and Rehabilitation
Volume 90, Issue 9 , Pages 1499-1505, September 2009

Article Tools