Population and methods

We conducted a prospective observational population based cohort study in a defined population aged 65 years and older in three urban areas in Norway (Stavanger, Trondheim, and Harstad) and their surrounding communities. The study lasted from 1 January 1990 to 31 December 1997 and involved 459 904 person years of observation. The average population during the study period was 57 488. The total number of person years of follow up was divided into exposure during the eight colder seasons (from 1 October through 31 March), and during the eight milder seasons (from 1 April through 30 September). The same subjects served both as the exposed population (exposure time during the colder seasons) and the control population (exposure time during the milder seasons). New subjects entered the population by reaching the age inclusion limit (65 years) or by moving into the study area. The subjects left the study population by emigrating or by dying. The information on population changes has been obtained from the Central Population Register of Norway that prospectively records individual information and is highly accurate.

Outcome was defined as a fall related fracture. Cases were identified through an ongoing prospective registration of all injuries performed by the National Injury Sample Register for Norway. The system registers injuries according to adapted classification and protocol for registration of injuries in the Nordic countries.¹⁹ The registration covers all medical facilities able to treat fractures in the study population. All fractures are registered, including those occurring among the institutionalized elderly people. Fractures caused by motor vehicle crashes and occupational injuries are excluded. Multiple fractures from the same injury event were calculated as a single case. Repeated fractures to the same person on different occasions were calculated as separate cases.

We stratified the analysis by age (65–79 years and 80 years and older), sex, and nature of injury. For nature of injury, cases were grouped as follows: hip fracture (International Classification of Diseases, ninth revision (ICD-9) code 820), arm fracture (ICD-9 codes 812–817), other fracture (ICD-9 codes 800–829 excluding hip and arm fractures). In the case of multiple fractures occurring in a single injury event, the following order of preference was followed: hip fracture, arm fracture, other fracture. The seasonal pattern was analyzed by calculating the relative risk for injury during the exposure to colder seasons compared with the exposure to the milder seasons. We also analyzed month-by-month changes in incidence rates, but that did not help to better clarify the seasonal patterns, due to smaller samples and larger random variations. Relative risk (RR) was calculated as the incidence rate ratio. Confidence intervals (95% CI) for the RR were calculated for each age, sex, and nature of injury subgroup by applying estimation method used for the person time denominator.²⁰ The joint incidence rate for both sexes within the same age and nature of injury subgroup was estimated by the Mantel-Haenszel procedure using the relative size of the groups as the weighting factors.²⁰

The contribution of icy and slippery conditions to the incidence of injuries was analyzed by classifying cases in those caused by slipping on ice and snow and those due to all other mechanisms. We calculated incidence rates stratified by each age, sex, and nature of injury.

Results

From 1990–97, 10 992 (2390 per 100 000 person years) fall related fractures occurred among people aged 65 years and older in the study population. Of these, 75% were caused by falls on the same level, 14% caused by falls from a level less than 1.5 m high, and fewer than 1% caused by falls from a level higher than 1.5 m. For 8% of the cases this information was missing.

Table 1 shows the incidence rate of injuries by nature of injury, season of the year, age, and sex.

There was a statistically significant difference in incidence rate of injuries occurring during the colder seasons for each sex and age subgroup. The difference was larger among people aged 65–79 years (RR = 1.39) than among people aged 80 years and older (RR = 1.17). It was similar for both men and women in the respective age subgroups. Approximately 16% of the cases in the age group 65–79 years and 8% of the cases in the older age group could be attributed to effects associated with colder season.

The difference in incidence rate of fractures between the colder and warmer seasons is most pronounced for arm fractures. Among people aged 65–79 years, the incidence rate of arm fractures was 69% (95% CI = 56% to 83%) higher during the colder season compared with milder season. The pattern was similar for women and men. Among people aged 80 years and older, the incidence rate of arm fractures was 30% (95% CI = 13% to 43%) higher during the colder compared with the milder season. The effect was higher among men (RR = 1.52) than women (RR = 1.24).

Twenty seven per cent (95% CI 15% to 37%) more hip fractures occurred during the colder than the milder season among people aged 65–79 years. The effect was somewhat higher among men (RR = 1.42) than women (RR = 1.20). Twelve per cent of all hip fractures occurring among people aged 65–79 years were attributed to the season effect. Among people aged 80 years and older, the seasonal pattern in occurrence of hip fractures was moderate (RR = 1.08, 95% CI 1.00 to 1.15). Only 4% of hip fractures were attributed to the effect of season.

Other fractures also showed statistically significant seasonal variation. Among people aged 65–79 years, the effect was smaller (RR = 1.12) than among people aged 80 years and older (RR = 1.34).

Table 2 shows the incidence rate of injuries by age, sex, nature of injury, season, and cause of accident. The excess incidence of hip fractures is entirely due to the fractures occurring by slipping on ice and snow. A similar finding is suggested for arm fractures. For other fractures, the excess risk is less clearly related to slipping on ice and snow.

Discussion

Studies investigating seasonal pattern in the incidence of hip fracture are contradictory.^1–13 In the current study, a seasonal variation in incidence of fall related fractures among elderly people exists for all types of fractures and in all age and sex groups. The seasonal pattern is most pronounced for arm fractures, and in the younger age group. Our study strongly suggests that the seasonal effect on the incidence of hip and arm fractures occurs entirely due to slipping on ice and snow. Seasonal effects are less likely to occur in areas with mild winter temperatures. This may explain the lack of seasonal variations in some studies. A seasonal effect on hip fractures among the oldest old is small and unlikely to be observed in studies performed on small samples.

Our study suggests that preventive measures should be targeted to icy and slippery surface conditions. Removing ice and snow and increasing friction of the icy surfaces with friction increasing materials are preventive measures that are centuries old. Such measures have, in various extents, been applied in our population during the study period. Population attributable fractions suggest a substantial prevention potential from a more intensive use of these measures. However, caution is required before translating the population attributable fractions calculated in this study into risk reduction estimates. The population attributable fraction is not the same as the preventable fraction.²¹ The fact that a proportion of injuries can be attributed to slipping on ice and snow does not imply that the same proportion of injuries would be avoided if all exposure to slipping on ice and snow was removed. An unknown number of injuries caused by slipping on ice would still occur because of competing risks. Removing ice and snow, or increasing stepping surface friction, could increase the exposure time for walking outdoors, as more people might feel safe to do so. Also, other mechanisms may play a part. From our data it is difficult to estimate how large the injury prevention effect of the improved ice and snow removal and the application of friction increasing materials would be. Intervention studies are required to answer that question.

This study has some limitations. Cases are missed if treated outside the registration system catchment area. This is, however, unlikely to cause major distortion of the findings. People are probably more likely to travel during the warmer season, leading to a differential underestimation of incidence rates for colder and warmer seasons. National hospital discharge statistics show that only about 1%–2% of hip fractures in our study population are treated in hospitals not represented in the injury register. Possible differential under-reporting of that size would not significantly influence the estimates. Further, our injury registration system is designed so that people injured and receiving initial treatment at other places, but coming for the follow up visit to a facility involved in the registration, are captured by the system. Information bias may also play a part. People may over-report by wrongly describing injuries as related to slipping on ice and snow, or under-report by failing to describe them correctly. Under-reporting is more likely and cannot be easily ruled out. But it would make our conclusions about the importance of slipping on ice and snow even stronger.

Finally, our findings are only valid for Norway. But weather in Norway is not much different from that of other Scandinavian countries, Central and Mid West states in the US, and many other countries in the world. It makes our findings likely to explain the seasonal variations in fractures that have been reported from many countries.

Implications for prevention

Slipping on ice and snow seems to be the main factor contributing to the excess seasonal risk of hip, arm, and other fractures among elderly people. Preventive measures targeting these causal mechanisms are likely to result in a reduced risk of fracture, but the size of the effect is difficult to estimate with certainty. A variety of preventive measures could be applied—health education, use of safety equipment, removal of ice and snow, or home food delivery services to the elderly during cold periods. Little is known about the costs and effectiveness of such measures, and good evaluation studies are needed to help select them.