2026 年 8 巻 3 号 p. 73-82
BACKGROUND
This study aimed to validate self-reported fractures in the Japan Multi-Institutional Collaborative Cohort (J-MICC).
METHODS
Of the 10,332 participants who were followed up for 10 years, 1,137 individuals with self-reported fractures were included. Among them, 201 individuals were randomly selected, and 192 individuals with available medical records were analyzed. A false-positive fracture was defined as a self-reported fracture without a clinical diagnosis or record. We calculated the positive predictive value (PPV) stratified by the number of years from injury to response. Multivariate logistic regression analysis was used to elucidate the factors influencing false-positive fractures and to calculate the odds ratios (ORs) with 95% confidence intervals (CIs). The dependent variable was false-positive fractures, and the independent variables included factors such as fracture site, age, sex, body mass index, education, multisite fractures, multiple fractures, and number of years from injury to response.
RESULTS
The average age at the time of injury was 69.6 years (range, 50–80 years), and 48 individuals (25.0%) were male. Fractures were reported in the spine (41 individuals, 21.4%), upper limbs (57 individuals, 29.7%), lower limbs (71 individuals, 37.0%), and other locations (23 individuals, 11.9%), respectively. The PPV was 85.4% (164/192 individuals; 95% CI, 79.7–89.7) and was >90% if the period from injury to response was within 3 years. False-positive fractures were significantly associated with a period from injury to response >3 years (OR, 3.46; 95% CI, 1.43–8.37).
CONCLUSIONS
elf-reported fractures are useful outcomes; however, recall periods >3 years may increase the risk of misclassification.
Self-reporting is commonly used in large epidemiological studies to gather subject information1–6) because checking medical records or reports is time-consuming and expensive. However, verifying self-reported fractures from medical records or reports is challenging for practical, ethical, and economic reasons. Therefore, assessing the validity of self-reported fractures is crucial.
Several studies have validated self-reported fractures in different regions, including the USA7–10), Europe11–15), Australia16), and Japan17) (Table 1). Nakamura et al.17) examined 133 participants (87 women, 46 men; average age, 72.4 years) from the Oguni cohort in the Japan Public Health Center-based Prospective Study who reported vertebral, upper limb, or hip fractures on a 15-year follow-up questionnaire survey and reported a positive predictive value (PPV) of 80.5% for agreement between self-reports and medical records. To the best of our knowledge, no other studies have validated self-reported fractures in Japan.
| Study | Country | Participant | Outcome | ||||
|---|---|---|---|---|---|---|---|
| Number (male/female) |
Age, years | Cohort | Agreement, % |
Period from injury to response, years | Factor influencing false positive fracture | ||
| Bush et al., 1989 [7] | USA | 23 (NA) | Range, 65–90 | The Florida Geriatric Research Program | 60.9% | >10 | NA |
| Beard et al., 1990 [8] | USA | 304 (0/304) | NA | Population-Based Epidemiologic Research | 84.2% | >15 | Number of years from injury to response |
| Nevitt et al., 1992 [9] | USA | 9704 (0/9704) | NA | Elderly white women with radiographic reports and medical records | 80.0% | NA | NA |
| Chen et al., 2004 [10] | USA | 6371 (0/6371) | NA | The Women’s Health Initiative Observational Study and Clinical Trials | 76.1% | NA | Being African American, living in Northwest USA, past smoker, high body mass index (≥30.0 kg/m2), history of osteoporosis, history of fracture (≥age 55), and falls in last 12 months (≥3 times) |
| Honkanen et al., 1999 [11] | Finland | 425 (0/425) | Range, 47–56 | The Kuopio Osteoporosis Risk Factor and Prevention Study | 78.8% | 10 | NA |
| Ismail et al., 2000 [12] | Europe | 510 (151/359) | ≥50 | The multicenter European Prospective Osteoporosis Study | 84.9% | 5.1 | NA |
| Hundrup et al., 2004 [13] | Denmark | 23376 (0/23376) | NA | The Danish Nurse Cohort Study on prevention of Osteoporosis and Atherosclerosis in Postmenopausal Women | 61.9% | 17 | NA |
| Siggeirsdottir et al., 2007 [14] | Iceland | 2255 (960/1295) | Range, 66–95 | The Age Gene/Environment Susceptibility-Reykjavik Study | 80.7% | 35 | Number of years from injury to response |
| Baleanu et al., 2020 [15] | Belgium | 829 (0/829) | Range, 60–85 | The FRISBEE Study | 85.2% | 11.1 | Age (60–75 years), high body mass index (≥25.0 kg/m2), education (<high school), and non-hip fracture |
| Ivers et al., 2002 [16] | Australia | 318 (NA) | NA | The Blue Mountains Eye Study | 65.4% | 10 | NA |
| Nakamura et al., 2011 [17] | Japan | 133 (46/87) | Mean, 72.4 | The Oguni cohort in the Japan Public Health Center-based Prospective Study | 80.5% | 15 | Non-vertebral fracture, number of years from injury to response, and male |
NA, not available
The Japan Multi-Institutional Collaborative Cohort (J-MICC) study is among the largest population-based cohort studies in Japan investigating the influence of lifestyle habits and genetic predisposition (genotype) on the development of lifestyle-related diseases, including cancer18–20). One outcome assessed in the Saga region cohort of the J-MICC study was self-reported fracture. These data, if accurate and have an appropriate PPV, can be used to analyze fracture risk factors. Therefore, this study aimed to validate the self-reported fractures in a cohort from the J-MICC study.
We conducted a prospective cohort study involving participants from the Saga region cohort of the J-MICC study, initiated between 2005 and 2007. The baseline survey details have been previously reported18–20). In summary, 61,447 residents aged 40–69 years in Saga City were invited to participate via mail, and surveys were conducted by telephone. A total of 12,078 individuals participated in the baseline survey.
This study examined the occurrence of fractures during a 10-year follow-up period from the baseline study. A questionnaire written in Japanese (Appendix 1) was mailed to all participants’ homes 10 years after the baseline survey to collect self-reported information on fracture history and location over the past 10 years. The questionnaire included the date of injury, fracture site, cause of fracture, and medical institution where treatment was received. There was only one reminder, and inquiries regarding omissions were made via phone. Of the 10,332 participants included in the 10-year study, 1,137 had self-reported fractures. Among them, 201 individuals with fractures were randomly selected, and pertinent information was collected from medical institutions. In this study, the primary outcome was PPV in participants with a history of fracture only, which did not allow for sample size calculation, which was consistent with that in a previous report21). Therefore, the sample size of this study was set to 1.5 times that of a similar cohort study by Nakamura et al.17), which investigated 133 Japanese individuals with fractures. Random selection was performed in one of the five districts surveyed using the RAND function to generate random numbers in Excel.
We contacted medical institutions to gather information on fractures for the selected 201 individuals, including the diagnosis, basis for diagnosis, and date of diagnosis (Appendix 2). In case of no response, three contact attempts were made. Participants who visited medical institutions were included. Finally, data was not available for nine cases, leaving 192 individuals for analysis (Fig. 1).
In the baseline survey of the J-MICC study, informed consent was obtained from participants to receive and complete a follow-up survey via mail regarding the incidence of various diseases. In addition, participants were asked to complete a follow-up survey via mail regarding fractures, and their responses were considered as consent to participate in the follow-up surveys in each field. The study participants agreed to contact the medical institutions gathering their data. The research protocol was approved by the ethics committees of the Saga University Graduate School of Medicine (approval No. 17-11) and Nagoya University Graduate School of Medicine (approval no. 253).
DATA COLLECTIONThe fractures were identified at the site of the first interview. A ‘confirmed fracture’ was defined as a fracture validated by medical records at the clinics, regardless of timing. A false-positive fracture was defined as a self-reported fracture but lacking diagnosis or records in the clinics. This study included all fractures in a relatively young and healthy population. Unlike previous studies, few fragility fractures were observed. Therefore, we focused on anatomical locations and classified them. Fracture site included the spine (i.e., cervical, lumbar, and thoracic vertebral fractures)7–17), upper limb fractures (i.e., clavicle, scapula, humerus, radius, ulna, carpal [wrist], metacarpal [palm], and phalangeal [finger] fractures)22), lower limb fractures (i.e., ilium, sacrum, pubis and ischium, femur, patellar, tibia, fibula, tarsus [ankle], metatarsus [forefoot], and phalanges [toes] fractures)22), and the other fractures.
Hip fractures and vertebral compression fractures are well-established indicators of osteoporosis and are associated with increased mortality23); thus, they were included in the analysis. In addition, although distal radius fractures are generally less strongly associated with mortality, they were also examined because they are included in the definition of fragility fractures in several large-scale epidemiological models, such as the QFracture study24). The years of injury were calculated based on the survey responses.
In cases of multisite fractures from a single injury (n = 10, 5.2%) and multiple fractures on different dates (n = 3, 1.6%), the fracture site was randomly selected.
Characteristics of the study participants, including age, sex, body mass index, and educational background (elementary or middle school, high school or vocational school, college, or graduate school), were collected during the baseline survey.
STATISTICAL ANALYSESAs a primary outcome, we calculated the PPV with 95% confidence interval (CI), stratified by the number of years from injury to response, using the following formula: (confirmed fracture)/(confirmed fracture + false-positive fracture). Based on previous studies7–17), we defined a PPV of 80% or higher as high and determined that it was reliable to use as an outcome measure.
As a secondary outcome, univariate logistic regression analysis was used to elucidate the factors influencing false-positive fractures and to calculate the crude odds ratios (ORs) with 95% CI. Subsequently, multivariate logistic regression analysis was used to determine the adjusted ORs with 95% CI. The dependent variable was false-positive fractures. Independent variables included potential factors such as fracture site15),17), age (<65 years and ≥65 years)15), sex (male and female)17), body mass index (<25.0 kg/m2 and ≥25.0 kg/m2)10),15), education (elementary or junior high school, high school or vocational school, and university or graduate school)15), multisite fracture (absent and present), multiple fracture (absent and present), and number of years from injury to response17) (≤3 years and >3 years); this accounted for potential predictors, acknowledging that apparent associations may exist despite non-significant differences in the univariate analysis. The reference category had an OR of 1, with ORs for other categories relative to this reference category.
Complete case analysis using case-wise deletion was performed to address missing data (number of years from injury to response, n = 1). All reported p-values were two-sided, and the level of significance was set at 0.050. Analyses were conducted using the JMP Pro software program, version 17 (SAS Institute, Cary, NC, USA) and R statistical software, version 4.5.0 (R Foundation).
A total of 192 individuals were analyzed. The average age at the time of injury was 69.6 years (range, 50–80 years), and 48 individuals (25.0%) were male. Regarding the fracture site, 41 individuals (21.4%) reported spine fractures, 57 (29.7%) had upper limb fractures, 71 (37.0%) had lower limb fractures, and 23 (11.9%) had other fractures.
VALIDITY OF SELF-REPORTED FRACTURESThe PPV between self-reported fractures and medical facility records (confirmed fractures) was 85.4% (164/192 individuals; 95% CI, 79.7–89.7). The PPV was >90% if the period from injury to response was within 3 years and >85% if the period from injury to response was <10 years (Table 2).
| From injury to response, years | Participants, n | False positive fracture, n | Confirmed fracture, n | PPV, % (95% CI) |
|---|---|---|---|---|
| All fractures | ||||
| 0 | 26 | 1 | 25 | 96.2 (81.2–99.3) |
| 1 | 58 | 5 | 53 | 91.4 (81.4–96.3) |
| 2 | 88 | 9 | 79 | 89.8 (81.7–94.5) |
| 3 | 115 | 10 | 105 | 91.3 (84.7–95.2) |
| 4 | 138 | 15 | 123 | 89.1 (82.8–93.3) |
| 5 | 147 | 16 | 131 | 89.1 (83.0–93.2) |
| 6 | 161 | 18 | 143 | 88.8 (83.0–92.8) |
| 7 | 170 | 19 | 151 | 88.8 (83.2–92.7) |
| 8 | 180 | 23 | 157 | 87.2 (81.5–91.3) |
| 9 | 186 | 25 | 161 | 86.6 (81.0–90.8) |
| 10 | 191 | 27 | 164 | 85.4 (79.7–89.7) |
| Spine fractures | ||||
| 0 | 9 | 0 | 9 | 100 (70.1–100) |
| 1 | 15 | 2 | 13 | 86.7 (62.2–96.3) |
| 2 | 22 | 2 | 20 | 90.9 (72.2–97.5) |
| 3 | 29 | 2 | 27 | 93.1 (78.0–98.1) |
| 4 | 34 | 3 | 31 | 91.2 (77.1–97.0) |
| 5 | 36 | 3 | 33 | 91.7 (78.2–97.1) |
| 6 | 38 | 3 | 35 | 92.1 (79.2–97.3) |
| 7 | 38 | 3 | 35 | 92.1 (79.2–97.3) |
| 8 | 40 | 4 | 36 | 90.0 (77.0–96.0) |
| 9 | 40 | 4 | 36 | 90.0 (77.0–96.0) |
| 10 | 41 | 5 | 36 | 87.8 (74.5–94.7) |
| Upper limb fractures | ||||
| 0 | 7 | 0 | 7 | 100 (64.6–100) |
| 1 | 15 | 1 | 14 | 93.3 (70.1–98.8) |
| 2 | 23 | 3 | 20 | 87.0 (67.9–95.5) |
| 3 | 31 | 3 | 28 | 90.3 (75.1–96.6) |
| 4 | 38 | 4 | 34 | 89.5 (75.9–95.8) |
| 5 | 40 | 4 | 36 | 90.0 (77.0–96.0) |
| 6 | 43 | 4 | 39 | 90.7 (78.4–96.3) |
| 7 | 47 | 5 | 42 | 89.4 (77.5–95.4) |
| 8 | 51 | 7 | 44 | 86.3 (74.3–93.2) |
| 9 | 55 | 9 | 46 | 83.6 (71.7–91.1) |
| 10 | 56 | 9 | 47 | 83.9 (72.2–91.3) |
| Lower limb fractures | ||||
| 0 | 7 | 1 | 6 | 85.7 (48.7–97.4) |
| 1 | 20 | 1 | 19 | 95.0 (76.4–99.1) |
| 2 | 29 | 2 | 27 | 93.1 (78.0–98.1) |
| 3 | 39 | 3 | 36 | 92.3 (79.7–97.4) |
| 4 | 47 | 5 | 42 | 89.4 (77.5–95.4) |
| 5 | 50 | 6 | 44 | 88.0 (76.2–94.4) |
| 6 | 59 | 8 | 51 | 86.4 (75.4–92.9) |
| 7 | 63 | 8 | 55 | 87.3 (76.9–93.4) |
| 8 | 66 | 8 | 58 | 87.9 (77.9–93.7) |
| 9 | 67 | 8 | 59 | 88.1 (78.2–93.9) |
| 10 | 71 | 10 | 61 | 85.9 (76.0–92.2) |
| Other fractures | ||||
| 0 | 2 | 0 | 2 | 100 (34.2–100) |
| 1 | 7 | 1 | 6 | 85.7 (48.7–97.4) |
| 2 | 13 | 2 | 11 | 84.6 (57.8–95.7) |
| 3 | 15 | 2 | 13 | 86.7 (62.2–96.3) |
| 4 | 18 | 3 | 15 | 83.3 (60.7–94.2) |
| 5 | 20 | 3 | 17 | 85.0 (64.0–94.8) |
| 6 | 20 | 3 | 17 | 85.0 (64.0–94.8) |
| 7 | 21 | 3 | 18 | 85.7 (65.4–95.0) |
| 8 | 22 | 4 | 18 | 81.8 (61.5–92.7) |
| 9 | 23 | 4 | 19 | 82.6 (62.9–93.0) |
| 10 | 23 | 4 | 19 | 82.6 (62.9–93.0) |
PPV, positive predictive value; CI, confidence interval.
The PPV was 87.8% (36/41 individuals; 95% CI, 74.5–94.7) in the spine, 83.9% (47/56 individuals; 95% CI, 72.2–91.3) in the upper limb, 85.9% (61/71 individuals; 95% CI, 76.0–92.2) in the lower limb, and 82.6% (19/23 individuals; 95% CI, 62.9–93.0) in other locations. The PPV was >85% within 10 years for spinal and lower limb fractures, >85% within 8 years for upper limb fractures, and >85% within 3 years for other fractures.
The PPV was 100% (3/3 individuals; 95% CI, 43.9–100) for hip fractures, 87.5% (35/40 individuals; 95% CI, 73.9–94.5) for vertebral compression fractures, and 87.1% (27/31 individuals; 95% CI, 71.2–94.9) for distal radius fractures (Table 3).
| From injury to response, years | Participants, n | False positive fracture, n | Confirmed fracture, n | PPV, % (95% CI) |
|---|---|---|---|---|
| Hip fractures | ||||
| 0 | 1 | 0 | 1 | 100 (20.7–100) |
| 1 | 2 | 0 | 2 | 100 (34.2–100) |
| 2 | 2 | 0 | 2 | 100 (34.2–100) |
| 3 | 2 | 0 | 2 | 100 (34.2–100) |
| 4 | 2 | 0 | 2 | 100 (34.2–100) |
| 5 | 3 | 0 | 3 | 100 (43.9–100) |
| 6 | 3 | 0 | 3 | 100 (43.9–100) |
| 7 | 3 | 0 | 3 | 100 (43.9–100) |
| 8 | 3 | 0 | 3 | 100 (43.9–100) |
| 9 | 3 | 0 | 3 | 100 (43.9–100) |
| 10 | 3 | 0 | 3 | 100 (43.9–100) |
| Vertebral compression fractures | ||||
| 0 | 9 | 0 | 9 | 100 (70.1–100) |
| 1 | 15 | 2 | 13 | 86.7 (62.2–96.3) |
| 2 | 22 | 2 | 20 | 90.9 (72.2–97.5) |
| 3 | 28 | 2 | 26 | 92.9 (77.4–98.0) |
| 4 | 33 | 3 | 30 | 90.9 (76.4–96.9) |
| 5 | 35 | 3 | 32 | 91.4 (77.6–97.0) |
| 6 | 37 | 3 | 34 | 91.9 (78.7–97.2) |
| 7 | 37 | 3 | 34 | 91.9 (78.7–97.2) |
| 8 | 39 | 4 | 35 | 89.7 (76.4–95.9) |
| 9 | 39 | 4 | 35 | 89.7 (76.4–95.9) |
| 10 | 40 | 5 | 35 | 87.5 (73.9–94.5) |
| Distal radius fractures | ||||
| 0 | 2 | 0 | 2 | 100 (34.2–100) |
| 1 | 7 | 0 | 7 | 100 (64.6–100) |
| 2 | 9 | 0 | 9 | 100 (70.1–100) |
| 3 | 13 | 0 | 13 | 100 (77.2–100) |
| 4 | 18 | 0 | 18 | 100 (82.4–100) |
| 5 | 19 | 0 | 19 | 100 (83.2–100) |
| 6 | 21 | 0 | 21 | 100 (84.5–100) |
| 7 | 23 | 0 | 23 | 100 (85.7–100) |
| 8 | 27 | 2 | 25 | 92.6 (76.6–98.0) |
| 9 | 30 | 4 | 26 | 86.7 (70.4–94.7) |
| 10 | 31 | 4 | 27 | 87.1 (71.2–94.9) |
PPV, positive predictive value; CI, confidence interval.
The factors influencing false-positive results are summarized in Table 4. In the multivariate logistic regression analysis, false-positive fractures were significantly associated with the period from injury to response >3 years (OR, 3.46; 95% CI, 1.43–8.37).
| Number (%) | False positive (%) | Crude | Adjusteda | |||
|---|---|---|---|---|---|---|
| OR (95% CI) | p value | OR (95% CI) | p value | |||
| Fracture site, n (%) | ||||||
| Spine | 41 (21.4) | 5 (17.9) | 1.00 (Reference) | 1.00 (Reference) | ||
| Upper limb | 57 (29.7) | 9 (32.1) | 1.35 (0.42–4.37) | 0.617 | 1.24 (0.36–4.27) | 0.739 |
| Lower limb | 71 (37.0) | 10 (35.7) | 1.18 (0.37–3.73) | 0.778 | 1.10 (0.33–3.68) | 0.881 |
| Others | 23 (11.9) | 4 (14.3) | 1.52 (0.36–6.32) | 0.568 | 1.53 (0.34–6.94) | 0.580 |
| Age | ||||||
| <65 years | 51 (26.6) | 7 (25.0) | 1.00 (Reference) | 1.00 (Reference) | ||
| ≥65 years | 141 (73.4) | 21 (75.0) | 1.09 (0.44–2.77) | 0.840 | 1.07 (0.41–2.80) | 0.897 |
| Gender | ||||||
| Male | 48 (25.0) | 10 (35.7) | 1.00 (Reference) | 1.00 (Reference) | ||
| Female | 144 (75.0) | 18 (64.3) | 0.54 (0.23–1.28) | 0.161 | 0.51 (0.20–1.33) | 0.169 |
| Body mass index | ||||||
| <25.0 kg/m2 | 134 (69.8) | 16 (57.1) | 1.00 (Reference) | 1.00 (Reference) | ||
| ≥25.0 kg/m2 | 58 (30.2) | 12 (42.9) | 1.92 (0.84–4.38) | 0.119 | 1.60 (0.66–3.90) | 0.302 |
| Education | ||||||
| Elementary or junior high school | 15 (7.8) | 3 (10.7) | 1.00 (Reference) | 1.00 (Reference) | ||
| High school or vocational school | 143 (74.5) | 19 (67.9) | 0.61 (0.16–2.37) | 0.479 | 0.61 (0.14–2.69) | 0.515 |
| University or graduate school | 34 (17.7) | 6 (21.4) | 0.86 (0.18–4.01) | 0.845 | 0.79 (0.15–4.21) | 0.781 |
| Multisite fracture | ||||||
| Absent | 182 (94.8) | 27 (96.4) | 1.00 (Reference) | 1.00 (Reference) | ||
| Present | 10 (5.2) | 1 (3.6) | 1.57 (0.19–12.88) | 0.676 | 1.24 (0.14–10.78) | 0.848 |
| Multiple fracture | ||||||
| Absent | 189 (98.4) | 28 (100) | ||||
| Present | 3 (1.6) | 0 (0) | ||||
| Number of years from injury to response | ||||||
| ≤3 years | 114 (59.7) | 10 (35.7) | 1.00 (Reference) | 1.00 (Reference) | ||
| >3 years | 77 (40.3) | 18 (64.3) | 3.17 (1.37–7.32) | 0.007 | 3.46 (1.43–8.37) | 0.006 |
J-MICC, Japan Multi-Institutional Collaborative Cohort; OR, odds ratio; CI, confidence interval.
a The dependent variable was defined as a false-positive fracture. Independent variables included potential factors such as fracture site, age, sex, body mass index, education, multisite fracture, multiple fracture, and number of years from injury to response; this accounted for potential confounding, acknowledging that apparent associations may exist despite non-significant differences in the univariate analysis.
A post hoc power analysis was performed for logistic regression (z-tests) using G Power software (Version 3.1.9.6, Germany). The number of years from injury to response was the only significant factor (OR = 3.42) in this study (Table 3), and the power was calculated as 0.99 with a significance level of 0.05 (one-sided).
We validated self-reported fractures in a cohort of the J-MICC study and observed a PPV of approximately 85%, with 88%, 84%, 86%, and 83% for the spine, upper limb, lower limb, and other sites, respectively. Additionally, the PPV was >90% when the period from injury to response was <3 years.
Previous studies reported PPV ranging from approximately 61%–85%, with 50%–90%, 40%–100%, 50%–100%, and 20%–100% for the spine, upper limb, lower limb, and other sites, respectively7–17). The variation in PPV among studies could result from a lack of statistical power due to subgrouping, the difference in age distributions (i.e., cognitive function) across study populations, lack of information in the medical institution, and/or ease of understanding the questionnaire (e.g., illustrations [Appendix 1]). Although the variation in PPV among studies was caused by multiple factors, our study demonstrated a generally consistent PPV (>80%) across different sites. Furthermore, hip and vertebral compression fractures are well-established indicators of osteoporosis and are associated with increased mortality23), while distal radius fractures are included in the definition of fragility fractures in several large-scale epidemiological models, such as the QFracture study24); thus, those with similarly high PPVs were examined in this study (Table 3). This suggests that all types of fractures, including minor fractures15), in the Saga region cohort of the J-MICC study can be used as an outcome measure.
Consistent with our findings, previous studies8),14),17) have reported an association of an increase in the number of years elapsed since injury with a decrease in PPV. In the J-MICC study cohort, the accuracy of self-reported fracture data was notably high, particularly when the time elapsed from injury to response was within 3 years. Although it depends on the contents of the recall, some previous investigators have also thought that the potential impact of recall bias changes after three years or more25),26). Regarding fracture recall, there is a possibility that self-reporting may be inaccurate, particularly when the time elapsed from injury to response is within 3 years; thus, caution is required not only in research but also in clinical practice.
In contrast to our findings, previous studies have reported that factors such as fracture type (e.g., non-hip and non-vertebral fractures), older age (≥55 years), male sex, high body mass index, lower education (<high school), and past smoking are related to false-positive fractures10),15),17). The causes of these discrepancies remain unknown; however, possible factors may include insufficient statistical power due to stratification by various fracture sites, decline in cognitive function due to aging, low levels of health consciousness related to smoking and obesity, and lack of understanding due to low education levels. Based on a previous report27), we calculated the power of the most striking variable and found it to be sufficient. However, we were unable to calculate separate sample sizes for each of the multiple independent variables. The recommended formula for the sample size is n = 100 + 50i, where i refers to the number of independent variables in the final model28). Although multisite fractures and multiple fractures have rarely been investigated in previous studies, they are considered important14). In this study, no significant differences were detected between multisite fractures and false-positive fractures, which may result from the small number of the fractures. Further investigations with sufficient statistical power are necessary to arrive at definitive conclusions.
LIMITATIONSThis study has some limitations. First, false-negative and true-negative fractures could not be assessed. In our study, no database-based fracture registration or mechanism for matching was available at the individual level. To identify false negatives, it was necessary to check the fracture records at orthopedic clinics throughout Saga City to determine whether the participant was among them. However, this approach was unrealistic. Although this study did not assess false-negatives, their rate would likely increase over time; further investigations are necessary regarding the changes in the risk of false negatives over time. Previous studies have reported false-negative fractures of 0.04%–6.50%7),11–13). Second, there may have been unmeasured predictors. Previous studies reported that ethnicity (i.e., African American), living place (i.e., northwest USA), history of osteoporosis, history of fracture, and falls in the last 12 months (≥3 times) were possible factors related to false-positive fractures10),15). Third, we did not ask participants to check for false positives. Therefore, participants could have been diagnosed at a different institution, which may have increased the incidence of false-positive fractures and led to lower PPV calculations. In this study, 192 participants (with a response rate: 96%) responded to the 201 inquiries. Furthermore, all 192 participants mentioned the medical institutions where they received treatment. No information was missing regarding the medical institution of any of the 192 participants. Therefore, no false-positive cases were caused by missing information regarding the medical institution. Fourth, our study participants were limited to residents aged 50–80 years living in the Saga Prefecture. Furthermore, random selection was conducted in one of the five districts surveyed, possibly leading to selection bias. However, few reports are available on the validation of self-reported fractures in Japan, whereas many studies have used self-reported fractures as an outcome, and these results have been applied clinically. Our findings improve the understanding of the validity of self-reported fractures.
In our J-MICC study cohort, we found an 85% PPV for self-reported fractures, with specific values of approximately 88%, 84%, 86%, and 83% for the spine, upper limbs, lower limbs, and other sites, respectively. When the period from injury to response was within 3 years, the PPV exceeded 90%. These findings suggest that all types of fractures in the Saga region cohort of the JMICC study can be used as outcome measures; however, recall periods >3 years may increase the risk of misclassification. These results should be interpreted with caution because of recall bias and unmeasured false negatives. Although this study focused on community dwellers aged 50–80 years living in rural areas, the results suggest that self-reported fractures may be used as an outcome measure in epidemiological studies, particularly when the time between injury and response is within three years. Our findings improve the understanding of the validity of self-reported fractures.
The authors have no conflicts of interest to declare.
This study was supported by grants from the Japan Orthopaedics and Traumatology Research Foundation (Grant Number 557), Grants-in-Aid for Scientific Research for Priority Areas of Cancer (No. 17015018), Innovative Areas (No. 221S0001), and the Japan Society for the Promotion of Science (Grant Numbers 16H06277, 21K11679, and 22H04923 [CoBiA]).
The performance of all procedures involving human participants in this study were in accordance with the ethical standards of the Institutional Review Board of Saga University School of Medicine (approval no. 17-11), Nagoya University Graduate School of Medicine (approval no. 253), and the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.
We thank all the contributors to the Japan Multi-Institutional Collaborative Cohort (J-MICC) study and the orthopedic surgeons in Saga City who cooperated in this research.
T.K. and M.H. conceived the study and contributed to data acquisition, analysis, interpretation, drafting, revision, and approval of the final manuscript version. Y.N., T.F., C.S., and K.T. contributed to critical revision of the article and approved the final version of the manuscript.
