European Journal of Epidemiology

, Volume 22, Issue 9, pp 631–639

Inaccuracy in self-report of fractures may underestimate association with health outcomes when compared with medical record based fracture registry

Authors

    • Icelandic Heart Association
  • Thor Aspelund
    • Icelandic Heart Association
  • Gunnar Sigurdsson
    • Landspitalinn University Hospital
  • Brynjolfur Mogensen
    • Landspitalinn University Hospital
  • Milan Chang
    • Landspitalinn University Hospital
  • Birna Jonsdottir
    • Icelandic Heart Association
  • Gudny Eiriksdottir
    • Icelandic Heart Association
  • Lenore J. Launer
    • National Institute on Aging
  • Tamara B. Harris
    • National Institute on Aging
  • Brynjolfur Y. Jonsson
    • University Hospital
  • Vilmundur Gudnason
    • Icelandic Heart Association
    • University of Iceland
Locomotor Disease

DOI: 10.1007/s10654-007-9163-9

Cite this article as:
Siggeirsdottir, K., Aspelund, T., Sigurdsson, G. et al. Eur J Epidemiol (2007) 22: 631. doi:10.1007/s10654-007-9163-9

Abstract

Introduction and objective Misreporting fractures in questionnaires is known. However, the effect of misreporting on the association of fractures with subsequent health outcomes has not been examined. Methods Data from a fracture registry (FR) developed from an extensive review of radiographic and medical records were related to self-report of fracture for 2,255 participants from the AGES Reykjavik Study. This data was used to determine false negative and false positive rates of self-reported fractures, correlates of misreporting, and the potential effect of the misreporting on estimates of health outcomes following fractures. Results In women, the false positive rate decreased with age as the false negative rate increased with no clear trend with age in men. Kappa values for agreement between FR and self-report were generally higher in women than men with the best agreement for forearm fracture (men 0.64 and women 0.82) and the least for rib (men 0.28 and women 0.25). Impaired cognition was a major factor associated with discordant answers between FR and self-report, OR 1.7 (95% CI: 1.3–2.1) (P < 0.0001). We estimated the effect of misreporting on health after fracture by comparison of the association of the self-report of fracture and fracture from the FR, adjusting for those factors associated with discordance. The weighted attenuation factor measured by mobility and muscle strength was 11% (95% CI: 0–24%) when adjusted for age and sex but reduced to 6% (95% CI: −10–22%) when adjusted for cognitive impairment. Conclusion Studies of hip fractures should include an independent ascertainment of fracture but for other fractures this study supports the use of self-report.

Keywords

QuestionnaireSelf-reportFractureFunctionRegistryAGES-Reykjavik study

Introduction

Reliability of questionnaire has been subject of studies for various different diseases with huge variations as recently reviewed [1]. Epidemiological studies often rely on participants’ self-report of fracture. It is therefore important to understand the accuracy of the responses and to what extent inaccuracy in answerers might affect the association of fracture with functional or other outcomes ascertained as sequela of fracture. Studies comparing results of questionnaires and information obtained by medical and radiological records have been inconsistent [213] with both over-reporting [3] and under-reporting [6, 14] that could lead to bias in results. In older persons, cognitive status may contribute to the accuracy of the fracture information but the only publications we identified that examine this problem found no effect of cognition on fracture report [2, 3]. However, there are relatively few studies available that have assessed both false-positive or false-negative characteristics of self-report [3, 9, 12]. Further, to our knowledge there are no studies that estimate the effect of misclassification on the association of fractures to health outcomes in old age, including functional status. We examined these issues in the context of a large population-based epidemiologic study that includes fracture questionnaire data and a fracture registry (FR) based on medical and radiological records. Specifically we estimate the percent of false positive and negatives for fracture history, identify factors associated with misreporting, and study the impact of misclassification on analyses of basic mobility, muscle strength, activity of daily living (ADL) and health-related quality of life (HQoL).

Material and methods

The cohort

Our study cohort is the first 2,300 participants in the Age Gene/Environment Susceptibility (AGES)-Reykjavik Study, which started in 2002 [15]. The AGES-Reykjavik Study is an extension of a population-based study from the original Reykjavik Study cohort. The study started in 1967 and has been previously described [16]. Briefly, it consists of repeatedly examined men and women who were residing in Reykjavik and nearby communities on 1 December 1966. The original invited cohort was 30,795 individuals and comprised 35% of the Icelandic population born during the period 1907–1935, with a response rate of 70% to the study. Participants in the AGES-Reykjavik Study were selected randomly from survivors of the Reykjavik Study and the age and gender distribution in the first 2300 is similar to the total surviving cohort [15].

Baseline examination, AGES- Reykjavik Study

All the participants in the AGES-Reykjavik Study completed a detailed health questionnaire that was administered in the clinic by a trained interviewer in a relaxed environment. This questionnaire included questions on fractures similar to those asked in The Study of Osteoporotic Fracture in Men but without all details [17, 18]. Performance measures of basic mobility and muscle strength function were acquired. The basic mobility measures we report include the “Timed Up and Go” test [19] and six meter walk [20, 21]. The maximal isometric muscle strength, handgrip- and knee extension strengths were measured using an adjustable dynamometer chair (Good Strength, Metitur, Palokka. Finland. www.metitur.com) [22]. ADL was measured by asking the participants’ how difficult it was for him/her to; dress, eat, have a bath, transfer out of bed/chair or walk from room to room. The score given was 0–15 with the highest score associated with ADL dependency. Poor outcome was defined as a score over 1 and included 10% of the participants. The HQoL was assessed by using the standardized EQ-5D measurement developed by the EuroQol group [23]. Those who scored over 75 were defined as having a poorer health quality.

Cognition was examined as previously described using a definition based on a consensus of cognitive impairment [15]. Informed consent was obtained and the study was approved by the National Bioethics Committee in Iceland (VSN 00-063) and the National Institute on Aging Intramural Institutional Review Board.

Attainment of fracture history: the fracture registry

The FR includes all participants of the Reykjavik Study still alive on 1 March 2002. Any fracture occurring between the date of entry into the Reykjavik Study in 1967 and the date of entry into the AGES-Reykjavik Study was recorded. The median follow-up time for the fractures was 31 years.

All patient records in Iceland, including records on fractures, are accessible through the medical system by use of a personal identification number. Since the late 1960s, all fractures treated on an out-patient basis in Reykjavik were referred to the only out-patient trauma clinic at the Landspitalinn University Hospital (previously the Reykjavik City Hospital). Fractures requiring in-patient treatment were referred directly to one of the three hospitals on call. The survey includes all hospitals in Reykjavik and Akureyri Hospital, the largest hospital outside Reykjavik. In addition to the patient records, we also searched for radiography records of fracture at a private radiology clinic servicing the GP practices in Reykjavik. All medical records, (including referral letter, comeback to the clinic, etc., in which fracture was mentioned) for all the participants were examined and all the fractures were verified. In cases where radiographs of rib- and vertebral fractures were not available, all medical records obtainable on the episode were examined by a single orthopedic surgeon who classified it as a fracture or not. Avulsion detachments less than 5 × 6 mm2, malignancy-caused fractures, and stress fractures were excluded. The FR contains only confirmed fractures, the nature of the trauma leading to the fracture, and the date of the fracture.

All medical records of participants who reported a fracture but who had not been identified as having a fracture by the AGES-Reykjavik FR were systematically re-examined with respect to whether a fracture had been missed. The registry had a 3% missing rate for a hip fracture, 2.7% for a forearm fractures and 4.1% for vertebral fractures. The AGES-Reykjavik Study FR has therefore a capture rate of about 97% for hip, forearm and clinical vertebral fracture. The medical records of 10% of participants or total 77, who did not report a fracture and wear not in the FR, were examined and no fracture identified. For analyses of the effect of fractures on basic mobility, muscle strength, ADL and HQoL all identified fractures were used.

Definitions

The FR was regarded as the reference standard when defining false positive, false negative, sensitivity, specificity, positive predictive value, and negative predictive value [2426]. False positives were defined as participants with positive self-report but a negative diagnosis according to the FR. False negatives were defined as the opposite. False positive rate is defined as 1-specificity and false negative rate as 1-sensitivity.

Sensitivity and specificity were defined as the proportion of true positives, respectively, true negatives fractures that were correctly identified by the participants according to the FR. Positive predictive value and negative predictive value was defined as the probability that the participant had correctly diagnosed a fracture or not.

Statistical analyses

The prevalence of individuals with fractures, cumulative over age groups, was compared according to self-report and FR. The McNemar’s test was used to determine if there was a statistical difference in the prevalence. To describe the agreement between the self-report and the FR by fracture types the kappa statistic was used. The same fracture was never counted twice in an individual and for a given site only the first fracture was used. The association between concordance/discordance and cognitive impairment was tested by logistic regression controlling for age and sex. The effect of fracture on basic mobility and muscle strength was estimated by a linear regression model with age and sex adjustment. The functional measures as response variables were adjusted for non- normality of distribution by using log-scale.

Additional analyses included adjustment for cognitive impairment. The results from the regression analyses are given as the percentage difference between fractured and non-fractured participants in performance on the functional tests. To examine if there was a consistent or constant difference between the results obtained by the two different methods in the data acquisition the relationship between the set of regression results estimated using self-report data versus FR data was compared by displaying them on a scatter plot with a weighted least squares trend line superimposed. The weights were the reciprocal of the standard error of the regression coefficient using the registry. We defined the attenuation factor from the slope of the trend line. It quantifies the average over- or underestimation of function with fracture for self-reports instead of FR.

Data analysis was conducted by using SAS software Version 9.1 (SAS Institute Inc., Cary, NC, USA).

Results

The cohort available for the fractures accuracy analysis consisted of 2,255 participants (Fig. 1). This sample consisted of 1,295 women (57%) with mean age of 76 years (range 66–95) and 960 men (43%) with mean age of 76 years (range 67–93). Evaluation of cognitive status was available for 2,094 participants, 895 men and 1,199 women. The age distribution in men and women was similar, (P = 0.08. Chi-square test 4df) (Table 1).
https://static-content.springer.com/image/art%3A10.1007%2Fs10654-007-9163-9/MediaObjects/10654_2007_9163_Fig1_HTML.gif
Fig. 1

Flow chart of the recruitment and number of participants according to sex

Table 1

Fracture information of the participants grouped by sex and age

Age group, years

Women

Men

<69

70–74

75–79

80–84

85+

Total

<69

70–74

75–79

80–84

85+

Total

n

174

351

343

338

89

1295

141

249

275

211

84

960

Reporting fracture, %

41.0

47.5

57.9

60.0

54.2

53.1

48.0

36.3

40.0

31.1

37.0

38.0

Fracture registry, %

34.3

40.2

55.2

58.1

56.6

49.2

35.4

29.2

27.3

26.9

32.1

29.3

False positive, %

12.2

11.0

10.1

9.2

8.6

10.4

14.3

15.1

19.9

13.6

14.7

16.0

False negative, %

5.5

4.3

7.4

6.0

11.4

6.1

2.4

7.8

7.5

9.7

5.9

7.1

Prevalence of fractures by self-report and registry

The number of individuals with self-reported fractures was consistently higher in all age groups for both genders compared with the FR except for the oldest women >85 years. According to the FR 40.9% of the participants sustained at least one fracture (49.2% women and 29.3%), compared to 46.8% from the questionnaire (53.1% women and 38.0% men). Among women both registered and reported fractures increased up to the age of 85 but in men there was no association with age (Table 1).

There was no significant difference in prevalence between FR and self-report for forearm, vertebral, hand and finger fractures (Table 2). The agreement between self-report and the FR as measured by the kappa statistics for both genders, varied from 0.11 to 0.80. Men answered consistently less reliably than women except for rib fractures. To compare the results obtained by the FR and self-report the sensitivity and positive predictive values of the self-report were calculated using the FR as a reference standard. There was considerable variation in sensitivity and positive predictive values of the self-report by the type of fracture, which were good for hip fractures combined (0.79 and 0.69, respectively, for both genders) and for forearm fractures (0.82 and 0.86, respectively, for both genders) but less so for other fractures. The specificity and negative predictive values were overall very good 0.90–1.00 and 0.95–0.99, respectively, for both genders.
Table 2

Percentage of individuals with fractures. Sensitivity, specificity, positive predictive value, negative predictive value and Kappa value of the questionnaire in 2,255 individuals

Fractures

Gender

Fracture registry

Self-report

Sensitivity

Specificity

Positive predictive value

Negative predictive value

Kappa value

Hip fracture*

Men

1.6

1.2

0.27

0.99

0.36

0.99

0.30

Women*

3.9

2.5

0.42

0.99

0.66

0.98

0.50

All*

2.9

1.9

0.39

0.99

0.58

0.98

0.45

Hip fracture combinedb

Men*

2.4

3.4

0.70

0.98

0.49

0.99

0.56

Women

6.3

6.0

0.82

0.99

0.78

0.99

0.78

Allb

5.1

4.4

0.79

0.98

0.69

0.99

0.73a

Vertebra

Men

3.9

4.2

0.54

0.98

0.58

0.98

0.54

Women

9.8

10.4

0.65

0.96

0.61

0.96

0.58

All

7.4

7.6

0.62

0.97

0.60

0.97

0.58

Forearm

Men

6.7

6.9

0.66

0.98

0.67

0.98

0.64

Women

27.0

26.0

0.85

0.96

0.89

0.95

0.82

All

18.8

18.9

0.82

0.97

0.86

0.96

0.80a

Rib*

Men*

6.1

14.2

0.57

0.89

0.24

0.97

0.28

Women*

4.7

10.4

0.48

0.92

0.22

0.97

0.25

All*

5.3

12.0

0.52

0.90

0.23

0.97

0.26

Shoulder and Clavicle*

Men*

3.0

7.0

0.69

0.95

0.30

0.99

0.39

Women*

9.1

11.2

0.74

0.95

0.60

0.97

0.62

All*

6.5

9.5

0.73

0.95

0.50

0.98

0.56a

Hand and Finger

Men

9.1

9.4

0.36

0.93

0.34

0.94

0.28

Women

7.6

8.4

0.53

0.95

0.48

0.96

0.46

All

8.2

8.8

0.45

0.94

0.42

0.95

0.38a

Ankle and Foot*

Men*

8.2

13.8

0.58

0.90

0.35

0.96

0.37

Women*

13.9

16.8

0.71

0.92

0.58

0.95

0.58

All*

11.5

15.5

0.67

0.91

0.49

0.96

0.50a

Pelvic

Men

0.7

0.6

0.29

1.00

0.33

1.00

0.30

Women

1.7

1.3

0.41

0.99

0.53

0.99

0.45

All

1.3

1.0

0.38

1.00

0.48

0.99

0.42

Femur*

Men*

0.0

2.2

0.00

0.98

0.00

1.00

0.00

Women*

0.5

3.5

0.67

0.97

0.09

0.99

0.15

All*

0.3

2.9

0.67

0.97

0.06

0.99

0.11a

*Difference in prevalence between fracture self-report and fracture registry P < 0.05

aKappa difference between men and women P < 0.05

bHip fracture, femoral shaft fracture and pelvic fracture

All medical records of participants who gave false positive responses according to the FR were systematically re-evaluated with respect to whether a fracture had been missed. The results for hip fractures are shown in Fig. 2. Two additional hip fractures were found, one participant had sustained a hip fracture outside Iceland and the other was not on the roster used for searching for fractures making the rate of missing a hip fracture 3% for the FR. The same value for forearm fractures was 2.7 and 4.1% for vertebral fractures.
https://static-content.springer.com/image/art%3A10.1007%2Fs10654-007-9163-9/MediaObjects/10654_2007_9163_Fig2_HTML.gif
Fig. 2

Distribution of participants with hip fracture into false positive and false negative self-report response group. Potential explanations for the discrepancies are given in the bottom panel boxes. SR self-report, FR fracture registry

False negative reporting rate varied considerably according to fracture type and was 61% for the hip where 67 participants sustained a hip fracture but only 27 of them correctly reported the fracture (Fig. 2). When examined more closely 36 of 40 individuals who did not report a hip fracture reported fractures in proximity to the hip (33 individuals femoral shaft fracture, 3 pelvis fracture, and 1 both fractures) instead of a hip fracture. These were therefore defined as reporting a hip fracture and a group of “hip fracture combined” created reducing the false negative report rate to 21% for the hip. For forearm, the false negative rate was 18 and 38%, for vertebral fractures but that was not explained by misreporting of fractures at similar site as we found for hip fracture. No additional fractures were identified when medical records were examined from 77 (10%) of participants who did not report a fracture or were in the FR.

Factors contributing to misreporting

In order to understand factors likely to contribute to incorrect answers the effect of age and cognitive impairment was analyzed as well as the time since the fracture occurred. Participants were asked about their age when they fractured their forearm and vertebra. The Spearman rank correlation between reported age and age in the FR was 0.85 for vertebral fracture and 0.82 for forearm fracture. Those who recalled their first fracture were fairly correct about when it occurred. The average difference in time for vertebra was 1 year (SD 5.5) and for forearm 0.7 year (SD 6.4). Participants who did not recall their vertebral fracture were at the time they answered the questionnaire on average 2.5 years older than participants who did recall their age at fracture (P = 0.0043 Wilcoxon). There was not a significant difference in age with respect to being able to recall the age of forearm fracture.

In order to estimate whether the differences between FR and self-report could be explained by cognitive impairment we analyzed false positive and false negative response with respect to the results from the cognitive diagnosis applied (Table 3). Impaired cognition was a major factor associated with discordant answers between FR and self-report, OR 1.7 (95% CI: 1.3–2.1) (P < 0.0001). The participants who gave false negative responses were more likely to be cognitively impaired compared to those who were concordant in their answers (P < 0.0001 all fractures, P < 0.0001 vertebra, P = 0.0002 forearm) (Table 3). After adjustment for age, sex and cognitive impaired it was found that participants are more likely to forget fractures that occurred 10 years or more [OR 1.43 (CI; 1.17–1.75), P = 0.006]. On the contrary, we did not find any significant difference for those who gave false positive answers.
Table 3

Percentage of participants with impaired cognition in three groups of; concordant, false positive and false negative results according to the fracture type

Fracture type

Concordance

False positive

False negative

Impaired cognition

Impaired cognition

Impaired cognition

n

%

n

%

P-value

n

%

P-Value

Alla

1,589

12.9

251

13.9

0.78

129

32.6

<0.0001

Hip

2,043

14.0

15

20.0

0.52

34

20.6

0.24

Hip combined

2,044

14.0

31

22.6

0.18

19

15.8

0.83

Vertebraea

1,942

13.2

63

19.1

0.19

51

39.2

<0.0001

Forearma

1,876

13.5

49

18.4

0.33

65

30.8

0.0002

aParticipants who reported a fracture before enrollment into the Reykjavik Study were excluded

Examination of possible consequences of misclassification

To study the possible consequences of differences between using the self-report and the FR the effect of fracture on mobility, muscle strength, ADL and HQoL were examined in the cohort. The hip fracture combined, vertebra and forearm were used and an effect on “Timed Up and Go”, 6 m walk and grip- and quadriceps strength was examined. Participants with fractures had consistently worse results compared to those without fractures according to both the FR and self-report. Figure 3 shows the summarized percent difference in performance of the three tests between fractured and non-fractured either classified by FR or self-report. The plotted regression results from the self-report against the FR results show that the self-report underestimated the effect of fractures on health outcome with an attenuation factor of 11% (95% CI: 0–24%) after an adjustment for sex and age. If, in addition, an adjustment was made for cognitive impairment the difference between using the self-report instead of FR was reduced to an average of 6% (95% CI: −10–22%). Similar results were obtained for ADL and HQoL (Supplement Figure 1). It is evident from Fig. 3 that there was a small discrepancy between the self-report and FR in forearm and vertebra but for the hip fractures combined the difference was larger. We found 22 participants who incorrectly reported fractures on femur or pelvis and when removed from the analyses the attenuation factor decreased to 9% (95% CI: −3–20) from 11%.
https://static-content.springer.com/image/art%3A10.1007%2Fs10654-007-9163-9/MediaObjects/10654_2007_9163_Fig3_HTML.gif
Fig. 3

Comparison of the fracture registry and self-report, association of hip, vertebra, forearm and all fracture to the Timed Up and Go, 6 m walk, grip- and knee extension strength. Adjusted for age and sex. Superimposed are a weighted least squares regression line and a 45° reference line. Scales show percentage difference in performance between fractured and non-fractured subjects using questionniare (y-axis) and registry (x-axis)

Discussion

Our study evaluated the accuracy of a fracture questionnaire by comparison with a FR based on medical and radiological records of participants in the AGES-Reykjavik Study, a study of people 66 years and older with a median follow up time of 31 years. Our detailed comparison between the two methods identified a number of false positive and false negative responses, which was to some extent explained by a cognitive impairment. The impact of fractures on function and HQoL was examined to evaluate any difference in the results from the two methods in identifying fractures used. This comparison demonstrated a small dilution in the effect when using the questionnaire results compared to using the registry for forearm and vertebra, but greater for hip fractures.

The small differences in the effect by forearm and vertebra fractures might be explained by non-significant differences between self-report and FR for these fractures, despite the fact that in most cases the FR had stronger association with diminished function. The fact that the results differed for hip fractures might partly be explained by the low prevalence of hip fractures in the study period, so the results were more sensitive to false positive responses, such as reporting pelvis and femur as a hip fracture due to anatomical misunderstanding. However, even allowing adjustments of the self-report by creating the group of “combined” hip fractures, the association of hip fracture with poor health outcomes still was diluted using the self-report of hip fracture versus the actual fracture cases. This result suggests that outcomes of hip fracture studies based on self-report cases may underestimate the importance of the fracture in decline in function. Just as important, for other fractures this detailed study supports the use of self-report in epidemiological studies.

The reference standard in our study was a FR verified by medical and radiographic records. We took steps to improve the accuracy of the diagnosis of fracture from these records since we know that hospital records of medical diagnosis as well as medical records are not always an accurate source of information on the type of fractures [27]. Studies comparing medical records and discharge abstracts have shown substantial discrepancy often leading to an incorrect diagnostic code [27, 28]. We found in a pilot project (data not shown) that 18% of fracture diagnoses obtained from a computerized hospital system had an incorrect diagnosis, similar to the findings of another fracture study [29]. To address this problem in our registry, we reviewed the participant’s medical records with respect to the fracture diagnosis making use of radiological information. Our FR turned out to have about 97% capture rate, which was shown by reexamining all medical records of all participants with false positive responses to the self-report. This is similar to what has been previously reported for another fracture database [14].

In most studies the main concern has been the high prevalence of false positive answers to fracture questionnaire. Our study has an overall relatively low false positive rate, 1–10% for both genders. The range of false positive answers in other studies is somewhat larger [3, 12, 13]. Three studies report no or low false positive rate for hip and forearm fractures [2, 8, 13] which is similar to our results while others report higher rates [3, 4, 9, 12, 29]. For vertebra we found the false positive rate to be 3%, which is similar to published results [7]. More of our findings are in accordance with other studies; those who recall their fracture were fairly correct about the year it occurred as validated by the FR. [4, 5, 7, 10, 12, 13], older participants answered as correctly as younger people [4, 5, 10, 12] and men answered consistently less reliably than women [12]. Studies based on self-report of fractures are subject to errors of recall [3] but the use of well-trained interviewers applying the questionnaire in our study instead of obtaining answers by mail may have resulted in more accurate answers. In some cases when people gave a false positive answer they were found to have in their medical records a bruise or other trauma at the site they reported the fracture (Fig. 2). It is unclear that even with more extensive questioning it would be possible to completely eliminate this response.

Our study is the largest study today examining false negative answers in detail. The rates were 18–61%. However, because the prevalence of fracture is low this is a small percentage of the entire study group, 0.2–4% and the effect of misclassification would be diluted. The false positive rate would be much more important because the number of cases of fractures may be small (particularly for hip) and the effect of misreporting would dilute the association with outcome, as we have shown, or potentially with risk factors for fracture. Comparison with other studies is difficult since they are different in their designs with a false negative range from 0 to 40% [2, 3, 5, 6, 1214, 29]. The very low or none false negative rate reported by two studies [3, 12] could be explained by the small subsets of subjects used in those studies to evaluate the under-reporting. In these studies, it may not be possible to correctly identify the true prevalence of false negative because of its low prevalence in the population.

However, to obtain valid results in epidemiological research correct classification of fractures is necessary. Despite our use of a face-to-face interview, a number of participants incorrectly reported a fracture or trauma in close proximity to the hip as a hip fracture. Of the 40 false negative hip fracture responses 36 reported a fracture of the femur shaft or of the pelvis. This has also been seen in previous reports [4, 9]. The discrepancy may be explained by a combination of the terminology used for various parts of the femur and hip and the lack of anatomical understanding of the hip region by the general population. Use of a manikin might help the participant to report their fractures more correctly [12]. This highlights the universal complexity of designing and applying questions in fracture research [9]. For this reason studies have combined positive answers for hip, pelvis and femur fractures in order to capture more hip fractures [4, 9, 10]. Although this method captures more true hip fractures our study shows it affected the effect of fracture on basic mobility, strength, ADL and HQoL.

There is a general tendency to under diagnose vertebral fractures [30] but few studies have addressed this question [7, 10]. In addition there are few studies concerning the accuracy of a fracture questionnaire where both false positive and false negative results are presented simultaneously [3, 5, 9, 12]. Despite methodological differences in the structure of the various studies the current study has considerable relevance to the understanding of possible limitations in using questionnaires in fracture research.

One study [7] reflected on why vertebrae fractures were reported less often compared to other fractures, despite direct evidence in medical records and suggested that these fractures were not reported to the patients. This might as well be due to cognitive impairment of the patients as shown in our study. It is evident that false positive and false negative response is present in this study and the false negative response partly may be explained by cognitive impairment. It is of interest to note that this is not the case for the hip, but that may reflect the fact that people sustaining a hip fracture have a high mortality [31, 32] and likely more so if they are cognitively impaired [33]. Our results indicate that the cognitive status of the participants does matter in the general population above 66 years of age when answering a fracture questionnaire and should wherever possible be examined. However, cognitive impairment does not totally explain the differences between the results obtained when only using self-report compared to using the FR.

It is of interest to compare our finding with results from studies dealing with different diseases and conditions. A recent review examined numerous studies on self-report and compared them to other methods evaluating diseases and diseases risk. This review did cast doubt on studies relying on self-report and showed a considerable variability in sensitivity and specificity between studies [1]. Our findings show variation in a sensitivity and specificity depending on what fracture we are examining and may reflect to some extent the nature of the questions. This supports previous recommendations, that self-report should be used with caution and whenever it is possible not to use the self-report as the only data sources but rather to investigate the possibility of the use of existing data sources to obtain external validation [1].

A drawback to our study may be that we did not include fractures prior to the start of the Reykjavik Study in 1966 but this time frame was not specified in the questionnaire when asking if the individuals had had a fracture. This was not as important for forearm or vertebra fracture as we asked age at fracture but did not include this information for hip fracture. We believe this would have a small impact, however, since these fractures usually occur late in life. Another limitation may be that some people have multiple fractures so there may be added impact on the function estimated for a single fracture site. We did not estimate this from our data.

The use of a questionnaire is a reasonably accurate method to establish the prevalence of forearm and probably clinical vertebral fractures in a general population cohort. The same seems to apply for the studies of the consequences of fractures such as the effect on basic mobility, muscle strength, ADL and HQoL although the self-report frequently underestimated the effect. However, hip fractures may be more mis-reported and use of a records based system to identify and confirm those fractures would be important.

Acknowledgments

This study was funded by NIH contract N01-AG-12100 and in part by the NIA Intramural Research Program., the Icelandic Parliament and the Icelandic Heart Association. Conflict of interest: none declared.

Supplementary material

10654_2007_9163_MOESM1_ESM.pdf (89 kb)
Comparison of the fracture registry and self-report, association of hip, vertebra, forearm and all fracture to the activity of daily living and health related quality of life. Adjusted for age and sex. Superimposed is a weighted least squares regression line and a 45 degree reference line. The scales shows the odds ratio of poor performance. (PDF 88 kb)

Copyright information

© Springer Science+Business Media B.V. 2007