Osteoporosis International

, 19:1797

Hip fractures cluster in space: an epidemiological analysis in Portugal

Authors

    • Laboratório de BiomateriaisINEB – Instituto de Engenharia Biomédica
    • Faculdade de Medicina, Serviço de Higiene e EpidemiologiaUniversidade do Porto
  • S. M. Alves
    • Laboratório de BiomateriaisINEB – Instituto de Engenharia Biomédica
    • ESTSP – Escola Superior de Tecnologia da Saúde do Porto
  • M. Barbosa
    • Laboratório de BiomateriaisINEB – Instituto de Engenharia Biomédica
    • Faculdade de Engenharia, Departamento de Engenharia Metalúrgica e MateriaisUniversidade do Porto
  • H. Barros
    • Faculdade de Medicina, Serviço de Higiene e EpidemiologiaUniversidade do Porto
Original Article

DOI: 10.1007/s00198-008-0623-1

Cite this article as:
de Pina, M.F., Alves, S.M., Barbosa, M. et al. Osteoporos Int (2008) 19: 1797. doi:10.1007/s00198-008-0623-1

Abstract

Summary

Using Portuguese hospital registers (2000–2002) we calculated age-standardized incidence rates of hip fractures. Spatial clusters of high incidence rates were found, with annual averages (per 100,000 inhabitants) varying from 154.4 to 572.2 and 77.3 to 231.5 for women and men, respectively. Geographic inequalities in the occurrence of hip fractures were also found.

Introduction

The aim of this study was to identify spatial patterns in the incidence of hip fracture in Portugal during the period 2000 to 2002.

Methods

From the National Hospital Discharge Register, admissions of patients (50 years of age or more) with low-energy hip fracture were selected. Age-standardized incidence rates in relation to the municipality of the patients’ place of residence were calculated. Empirical Bayes estimators were used to smooth the local risk and spatial statistics methods were used to identify spatial clusters.

Results

Of 25,634 hip fractures in individuals aged 50 years or more caused by low or moderate impact, 19,759 occurred in women (age, mean±SD, 80.6±8.6 years) and 5,875 in men (age 77.7±10.0 years). Incidence rates increased exponentially with age, being higher in women nation-wide (female to male ratio from 1.5 to 5.1). Significant geographic differences were found: the incidence rates (95% CI) varied from 154.4 (153.6–155.3) to 572.2 (569.5–575.0) in women and 77.3 (76.64–78.05) to 231.5 (229.9–233.0) in men per 100,000 inhabitants. Spatial autocorrelation values (Moran index) were 0.56 and 0.45 for women and men, respectively.

Conclusion

Spatial clusters (p<0.0001) of high incidences were identified. Geographic differences in incidence rates were about threefold. Some regions had incidence rates as high as some north European countries. The geographic inequalities could be due to environmental or socioeconomic factors, but further investigation needs to be done to confirm this hypothesis.

Keywords

Geographic information systemsHip fracturesOsteoporosisSpatial analysisSpatial clustersSpatial dependency

Introduction

Osteoporosis is a well-recognized consequence of ageing but the intrinsic process of ageing is not sufficient to lead to the disease. Genetic, environmental and socioeconomic characteristics play an important role in the pathological process [1].

Fractures are the most visible and dramatic consequence of osteoporosis [2] and occur as a result of moderate or minimal trauma [3]. They are more frequent in women than in men. The incidence rates increase with age and tend to occur mainly in the proximal femur (hip), vertebrae and distal forearm (wrist) [2]. Hip fractures are the most serious, resulting in hospital admission and tending to recur. They are strongly related to low bone mineral density (BMD) and are regarded as surrogates for (severe) osteoporosis [4].

Worldwide the overall incidence of osteoporotic fractures has doubled in the past 25 years [5], partly reflecting the increased life expectancy and it constitutes an orthopaedic epidemic and a major public health problem. The geographic variation in the incidence of osteoporotic fractures, between and within countries, strongly supports a role for environmental factors [1, 6, 7, 8]; however spatial patterns and spatial clusters of hip fracture have not been studied in depth.

To explore the burden of osteoporotic hip fractures in Portugal, hospital admissions from 2000 to 2002 were analysed using a nation-wide hospital database with the aim of mapping, describing, quantifying and evaluating spatial patterns in the incidence of hip fractures in Portugal.

Materials and methods

Study area

The geographic area of study was continental Portugal, which in 2001 had a population of 9,869,343 inhabitants, distributed in 278 municipalities (mean±SD population of 35,501±56,918 inhabitants). The smallest municipality had 1,924 inhabitants while the largest (Lisbon) had 564,657 inhabitants. The municipalities are aggregated to form 18 districts. The area of Portugal is 92,391 km2 and the climate is Mediterranean temperate, with an average temperatures of 13°C in the north and 18°C in the south. In terms of sun exposure, the south of the country (Alentejo and Algarve) has the highest percentage of sunny days per year.

The majority of the population live in urban or semiurban areas (86%), which are distributed mainly along the coast. Figure 1 shows the Portuguese districts and the distribution of urban and rural areas.
https://static-content.springer.com/image/art%3A10.1007%2Fs00198-008-0623-1/MediaObjects/198_2008_623_Fig1_HTML.gif
Fig. 1

Distribution of urban, semiurban and rural areas in Portugal

Study population

Data from the National Hospital Discharge Register were used. In Portugal access to the national health-care system is universal and free for all the population from all social groups and all ages. Therefore, due to the high costs involved, hip fractures are almost always treated in public hospitals, and for that reason they are highly documented and the National Hospital Discharge Register records the totality of admissions with a diagnosis of hip fracture nation-wide. These data are updated every month by all public hospitals, and the database is managed by the Health Informatics and Financial Management Institute (Instituto de Gestão Informática e Financeira da Saúde – IGIF) of the Portuguese Health Ministry. No such data were available for the Portuguese archipelagos of the Azores and Madeira, and therefore they were not included in this study.

In the National Hospital Discharge Register, each record corresponds to one admission and contains information including: gender; age; first cause of admission (and up to 19 secondary causes), coded according to the International Classification of Diseases, version 9, Clinical Modification (ICD9-CM); main diagnosis (and up to 19 secondary diagnoses) also coded according to the ICD9-CM; clinical interventions (up to 20); surgical interventions; hospital providing the care; outcome (for example, discharge home, discharge to another hospital, deceased); number of days of hospitalization; geographic unit of the patient’s place of residence; and hospitalization costs.

We selected all admissions from 2000 to 2002 of individuals aged 50 years or more with a discharge diagnosis of hip fracture according the ICD9-CM classification (codes ICD9-CM 820.x) caused by a low or moderate fall. Bone cancer, readmissions for orthopaedic after-care or complications of surgical and medical care (codes ICD9-CM: V54.x and 996.4) were not considered.

Population denominator

We used the population count for municipalities by gender and 5-year age groups from the Portuguese Demographic Census of 2001, which is the middle of the study period.

Data analysis

Geographic information systems (GIS) and spatial statistical techniques were used to analyse the data and map the results. Each hospital admission was georeferenced according to the municipality of the patient’s place of residence. For each municipality the data were then aggregated by gender and 5-year age groups. The gender incidence rates were standardized by age using the direct method (Portuguese population of 2001 as the standard population).

Because the study involved small geographic areas (municipalities), some only with hundreds of inhabitants and few cases, the problem of small numbers [9] was accounted for using the empirical Bayes (EB) approach to “smooth” the local risk. This approach is a statistical estimation based on observed data and on prior knowledge about the parameters of interest [10]. The degree of “smoothing” reflects the level of confidence in the local observed risk, and it is a function of the size of the population and the variability of the incidence rates. The magnitude of the adjustments increases as the population of an area decreases. In areas with a small population (consequently with highly unstable incidence rates) the observed incidence rates are reduced to the average incidence rate of the neighbouring areas to produce an estimated incidence rate. In areas with large populations the confidence in the observed incidence rate is higher and the estimated incidence rate will be similar to the observed rate [10]. The estimated incidence rate reflects better the true risk in an area, because the effect of an artificially high observed incidence rate caused by a few cases in a small population is eliminated.

After adjusting the incidence rates by the EB method, the Moran index was computed in order to measure the spatial autocorrelation, that is the correlation between incidence rates in different municipalities [10]. A first order neighbourhood relationship was defined by the sharing of common boundaries between municipalities. The Moran index is a global indicator of autocorrelation and is useful to characterize a region of study. It provides a single value for the all sets of data. The interpretation of the index is similar to the interpretation of the coefficient of correlation r in a linear correlation. A value close to zero means that there is no autocorrelation, and the events occur randomly in space. A value close to 1 or −1 means there is a strong (positive or negative) autocorrelation and indicates that there is a spatial dependency in the occurrence of the disease, that is what happens in one area is correlated with what happens in the neighbouring areas, and the events do not occur randomly. A positive autocorrelation means that the values encountered in one area are similar to those in the adjacent areas, whereas a negative autocorrelation means that if one area has a high incidence rate the adjacent areas have low incidence rates, or if one area has a low incidence rate the adjacent areas have high incidence rates, compared to the average of the region.

However, when dealing with a great number of areas, it is possible that different spatial associations occur; therefore, one single value would not represent the underlying patterns. To deal with these different spatial associations the local Moran index, known as the local index of spatial autocorrelation (LISA) [11], was calculated. The LISA indicates the presence of spatial dependence in some areas, that is areas where the incidence rates are significantly correlated with the incidence rates of their neighbours. The municipalities were classified and mapped into four classes: high–high (areas with a high rate surrounded by other areas with a high rate), low–low (areas with a low rate surrounded by other areas with a low rate), low–high (area with a low rate surrounded by areas with a high rate) and high–low (the inverse of low–high).

Results

The database comprised a total of 36,846 hip fracture episodes, 12,892 (35%) in men (54.9±27.4 years old) and 23,954 (65%) in women (76.6±16.2 years old). Of the total admissions, 273 (less than 1%) were disregarded because they could not be geographically referenced as a consequence of poor quality information on the place of residence of the patients. There were 25,634 admissions for hip fracture in individuals aged 50 years or more caused by a low or moderate impact. These admissions comprised 19,759 (77%) women and 5,875 (23%) men aged 80.6±8.6 and 77.7±10.0 years, respectively, of whom 81% and 79%, respectively, were georeferenced to urban or semiurban areas.

The three most frequent causes of admission were ICD9-CM E888 “other and unspecified fall” (80.7%), E885 “fall on same level from slipping, tripping, or stumbling” (8.4%), and E887 “fracture, cause unspecified” (3.5%).

The incidence of hip fractures was higher in women than in men, the annual average crude rate being, respectively, 351.87 and 129.39 per 100,000 inhabitants. Figure 2 shows the exponential growth in annual average crude rates by age group and sex, and Table 1 presents the corresponding number of cases, population and incidence rates by gender and age group as well as female to male ratios of incidence rates.
https://static-content.springer.com/image/art%3A10.1007%2Fs00198-008-0623-1/MediaObjects/198_2008_623_Fig2_HTML.gif
Fig. 2

Incidence of hip fractures, by gender and age group

Table 1

Population by age group, total number of fractures, population from 2001 census, annual average incidence rates per 100,000 person-years (95% CI) and gender ratio

Age group (years)

Number of fractures

Population 2001

Annual average incidence rates (CI)

Gender ratio (female:male)

Women

Men

Women

Men

Women

Men

50–54

151

154

326,207

303,556

15.4 (11.4–20.2)

6.9 (12.5–22.1)

0.9

55–59

270

208

296,214

264,017

30.4 (24.4–37.3)

26.3 (20.3–33.1)

1.2

60–64

502

271

287,736

251,999

58.2 (49.3–66.9)

35.8 (28.7–43.9)

1.6

65–69

1,075

559

286,482

239,466

125.1 (112.1–138.0)

77.8 (66.5–89.1)

1.6

70–74

2,377

760

251,032

192,616

315.6 (293.6–337.6)

131.5 (115.2–147.5)

2.4

75–79

3,892

1,135

199,973

140,564

648.8 (613.4–684.0)

269.2 (242.0–296.3)

2.4

80–84

4,441

1,176

122,842

74,306

1205.1 (1143.7–1266.4)

527.5 (475.3–579.8)

2.3

85–89

4,305

1,022

70,681

35,383

2030.3 (1925.2–2135.3)

962.8 (860.6–1065.1)

2.1

90+

2,746

590

30,666

11,599

2984.8 (2791.4–3178.1)

1695.6 (1458.8–1932.8)

1.8

The age-standardized incidence rates (annual average), in relation to gender and the municipality of the patient’s place of residence, were adjusted by the EB method and the differences among regions, ranged from 154.4 to 572.2 (per 100,000 inhabitants) among women and from 77.4 to 231.5 (per 100,000 inhabitants) among men. Figure 3 shows the spatial distribution of the annual average of age-standardized incidence. The classes for the maps were generated using the quartile distribution of the incidence rates.
https://static-content.springer.com/image/art%3A10.1007%2Fs00198-008-0623-1/MediaObjects/198_2008_623_Fig3_HTML.gif
Fig. 3

Geographic distribution of age-standardized incidence rates, adjusted by the EB method

Large differences were not encountered when the age-standardized incidence rates were calculated in relation to type of area. The incidence rates (per 100,000 inhabitants) for urban, semiurban and rural areas were, respectively, 125.3, 119.1 and 105.5 among men and 372.8, 386.5 and 320.3 among women. These differences are statistically significant (p<0.0001).

The Moran indexes of spatial autocorrelation were 0.56 for women and 0.45 for men. Clusters of high and low rates were identified when the LISA was calculated. The resulting LISA maps are shown in Fig. 4.
https://static-content.springer.com/image/art%3A10.1007%2Fs00198-008-0623-1/MediaObjects/198_2008_623_Fig4_HTML.gif
Fig. 4

Spatial clusters of age-standardized incidence rates by municipality

The female to male ratio of adjusted incidence rates was also calculated for all the municipalities, and the results (Fig. 5) showed a geographic variation from 1.5 to 5.1, with the higher ratios in the south (Algarve) and central coastal area (from Lisboa to Coimbra) and the lowest in centre-south (Alentejo).
https://static-content.springer.com/image/art%3A10.1007%2Fs00198-008-0623-1/MediaObjects/198_2008_623_Fig5_HTML.gif
Fig. 5

Geographic distribution of female to male ratios of incidence rates

Discussion

The main objective of this work was to identify the spatial patterns of hip fractures in Portugal, and the results showed significant nation-wide geographic differences in the incidence rates. The geographic differences in age-standardized incidence rates, at municipality level, among women were 3.7-fold and among men were about 3-fold. In contrast to these high geographic differences, the differences in age-standardized incidence rates between rural and urban areas were smaller, being less than 20% higher in urban compared to rural areas, for both women and men. There are several studies comparing hip fractures in urban and rural areas, and almost all have shown higher incidence rates in urban areas [12, 13]. Although true also in Portugal, these differences do not explain the geographic pattern of hip fracture incidence in the country. Some urban areas, such as the Porto district (the second most important urban area in Portugal), had some of the lowest incidence rates, while the Bragança district, which is mainly rural, had the highest incidence rates.

No correlation was found between the pattern of hip fracture incidence rates and exposure to sunlight. The Algarve, in the south, that has the highest temperatures and the greatest number of sunny days per year, had hip fracture incidence rates among the highest. On the other hand regions such as Vila Real, in the north, or Castelo Branco in the centre, which have fewer sunny days per year and lower temperatures, had hip fracture incidence rates among the lowest.

There was a slight difference in the geographic pattern between genders. In men higher rates were found in the northeast, in some central municipalities and in the coastal south, and lower rates were found in the northwest, central east, and in some southern inland municipalities. In women higher rates were found in the northeast, all the coastal strip, except the northwest and part of central south (Alentejo), and lower rates were found in the northwest (Viana do Castelo, Porto and Aveiro districts), and many central inland municipalities. There was no evidence that the gender differences in the incidence of hip fractures were related to differences in accessibility of the health-care system in different regions. In Portugal access to the health-care system is free and universal. Different patterns of care between genders have been described for other diseases (e.g. coronary heart disease), but there are no reasons to believe that this is the case for hip fractures. This study used hospital admissions, georeferenced to the municipality of the patient’s place of residence. Despite transfer of care between hospitals in different regions to access better technological facilities, the patients were always georeferenced to their place of residence, and therefore the incidence rates reflect the local risk. The authors believe that differences in care between genders could be associated with differences in case fatality, but not in incidence rates. The reasons for gender differences may be related to occupational exposure and differences in risk of falls, although these aspects were not studied in this work. Further investigation is needed to better understand the regional differences between men and women in the incidence rates of hip fractures.

The values of the Moran index for spatial autocorrelation were high and significant, in both genders showing a positive spatial autocorrelation. This means that nearby areas tend to have similar the incidences of hip fracture. However, the incidence rates in women showed a higher Moran index, indicating a stronger spatial autocorrelation. As shown by statistically significant local spatial autocorrelation values (Fig. 4), there were clusters of higher incidence rates in the northeast around Bragança, as well as in some municipalities around the central coastal area (Setúbal). However, for women, a higher number of municipalities showed a significantly higher spatial correlation coefficient in this area. For men, clusters of high incidence rates were also found in the central southern area and the coastal southern area (Alentejo). For women, clusters of high rates were found along all the municipalities of the Algarve (south of Portugal). Clusters of lower rates for both genders were found in the northwest around Viana do Castelo. Clusters of lower rates were also found in some northern municipalities of Porto, Aveiro and Guarda, as well in some central municipalities around Santarém.

According to population projections for Portugal by the National Institute of Statistics (http://www.ine.pt), the number of people over 50 years of age in 2020 will be 4,329,624, with 2,360,952 women and 1,986,672 men, and in 2050 will be 4,669,189, with 2,502,614 women and 2,166,575 men. This means an increase in the population over 50 years old of about 871,000 individuals in 2020, and about 1,210,000 individuals in 2050, compared to 2001. Considering the number of osteoporotic hip fractures occurring in the study period, and making a proportional estimate, if the crude incidence rates were only due to ageing, in Portugal there would be around 10,877 osteoporotic hip fractures, by 2020 and 11,609 by 2050, an increase of 27% and 36% respectively, compared with the study period.

The results of this study are important in providing for the first time at the national level a clear understanding of the situation with regard to the occurrence of hip fractures in Portugal. As part of the Mediterranean Osteoporosis Study (MEDOS), the incidence of hip fractures reported for a small region in Porto district (less 10% of all the Portuguese territory) was 259.0 for women and 114.0 for men (per 100,000 inhabitants) [14]. In order to compare the results of this study with the MEDOS study, we calculated the age-standardized incidence rates of hip fractures for the same region around Porto city and found rates of 352.4 and 138.5 (per 100,000 inhabitants), respectively. Although these results indicate that the incidence rates increased 36% for women and 21% for men between 1988–1989 and 2000–2002 there are some differences in the methodologies between the two studies and therefore the comparison has to be made with caution. Because of the use of EB estimators to adjust the incidence rates the results were more realistic. The high incidence rates resulting from the small population problem, which do not reflect the actual risk of an area, were smoothed to a local rate.

In order to compare our results with those of other European studies, we also calculated the crude incidence rates for Portugal, and selected similar studies from other European countries. The criteria for selection of the studies were: use of hospital admissions with a diagnosis ICD9 820.x, patients aged 50 years and above, and fractures caused by low-energy impacts. In Portugal, the mean age of women who had a fracture was lower than in Norway (82.1±9.3 years) [15], Switzerland (82.0 years) [16], a region in northern Spain (84.0±8.0 years) [17] and Denmark (81.2±9.0 years) [18]. The mean age of men was lower than in Spain (78.0±12.0 years) [17] and Denmark (79.8±10.0 years) [18], and higher than in Norway (76.6±12.8 years) [15]. The overall annual average crude incidence rate in the study period among women was 351.87 per 100,000 inhabitants, which (comparing equivalent studies in Europe) is higher than in Germany (291.3) [19], and lower than in Spain (388.6) [17], Greece (407.57) [20], Norway (1180.0) [15], Denmark (598.0) [18] and Finland (467.0) [21]. A similar pattern was seen among men. The overall annual average crude incidence rate in the study period among men was 129.39 per 100,000 inhabitants, which is higher than in Germany (110.2) [19] and Spain (100.4) [17], and lower than in Greece (203.07) [20], Norway (440.0) [15], Denmark (284.0) [18] and Finland (233.0) [21].

The overall female to male ratio of incidence rates was 2.7, which is similar to that in other European countries. More than 50% of the municipalities had a female to male ratio between 2.5 and 3, and there was a geographic variation in the ratio, the minimum being 1.5 and the maximum 5.1. The reasons for these findings are not clear, and further investigation is needed.

Despite the importance of calculating overall crude incidence rates in order to compare results between countries, there were significant differences within the country. In Fig. 3 are displayed the annual average hip fracture incidence rates for men and women by municipality. These are the estimated rates after smoothing to the neighbourhood average rates. The incidence rates in women were higher than in men in all municipalities and had a higher variation. While the incidence rates (95% CI) among men varied 3.0-fold from 77.4 (76.64–78.05) to 231.5 (229.9–233.0) per 100,000 inhabitants, the incidence rates among women varied 3.7-fold from 154.4 (153.6–155.3) to 572.2 (569.5–575.0) per 100,000 inhabitants. Despite the fact that Portugal has, in general, lower rates than north European countries, the more localized analysis showed that municipalities with a higher incidence of hip fractures in Portugal had rates similar to those reported in Finland or Denmark [18, 21].

The incidence of low-energy trauma hip fractures in individuals over 50 years of age was much higher in women than in men in all regions as expected [3]. The female to male ratio was higher than 1 in all age groups, except in the age group 50 to 54 years, and showed an increase with age starting from 0.9, to a maximum of 2.4 in the age group 75 to 79 years. From this age group onwards, the ratios start to decrease, reaching a value of 1.8 in those more than 90 years of age.

Osteoporosis is not only the result of biological and genetic factors, but also of territorial occupation and socioeconomic and environmental conditions that together form the totality of external elements that influence the health and welfare of a population. Incorporation of the spatial view in hip fracture incidence studies makes an important contribution to elucidating the health–disease process and can lead to results different from those obtained in studies that do not consider the variable space. Our results showed that some external factors influenced the occurrence of hip fractures. Some studies have shown that socioeconomic or environmental factors can be related to differences in hip fracture incidence rates. In this study important geographic differences were identified in Portugal, but the reasons for these differences need to be clarified in further studies. These differences are not related to differences between rural and urban areas, and they do not seem to be related to sunlight exposure. The authors are developing further studies to correlate the distribution of hip fractures incidence with drinking water quality. Socioeconomic conditions, particularly in the past, could also be related to differences in hip fracture incidence. This needs further investigation.

The use of GIS allowed a fine description of hip fracture occurrence in Portugal and, together with spatial statistics, the identification of significant geographical clusters, creating the basis for a better understanding of the occurrence of hip fractures and providing new knowledge to inform decisions on preventive actions. Osteoporosis, and the consequent hip fractures, are considered an important problem in public health. However, they can only be prevented if their real intensity nation-wide is known..

The limitations of this study are concerned with the quality of the database. One problem is the fact that registers do not record a code for each patient, but instead a code for each admission. Because of this, it is impossible to separate new cases from cases involving readmission. However, this problem was avoided by excluding admissions for treatment and complications after an orthopaedic surgical intervention. Another problem that may have influenced the quality of the results was the fact that most registers were coded ICD9-CM E888 “Other and unspecified fall”. In this investigation we considered that all of them were low impact accidents, but some may not have been. We could not assess which factors, including environmental and social, contributed to the geographical pattern encountered in the country. Concerning the different patterns for the distribution of hip fractures in both sexes, especially in the south with higher rates, that is not related to different age distribution, information on the eventual clustering of different risk factors is much needed.

Acknowledgements

The authors acknowledge the support of the Portuguese Foundation for Science (grant no. POCI/SAU-ESP/58605/2004).

Conflicts of interest

None.

Copyright information

© International Osteoporosis Foundation and National Osteoporosis Foundation 2008