Secular reduction of excess mortality in hip fracture patients >85 years

Data from three earlier incidence studies on hip fracture in Oslo were used [14‐16]. These studies include all patients ≥20 years with hip fracture in the two-year periods 1978 to 1979 (n=2067) and 1988 to1989 (n=2697), and in the one-year period from 1st of May 1996 until 30th of April 1997 (n=1290). Hip fractures (International Classification of Diseases, ninth revision (ICD-9) code 820.X) were identified through diagnosis registers, operating theatres protocols, medical records, and x-ray records. Fractures in patients residing outside Oslo or fractures due to malignancy were not included in these studies.

Hip fractures in patients aged <50 years are infrequent, and the younger hip fracture patients differ considerably from the older hip fracture patients regarding comorbidity [17, 18]. For this reason, patients aged <50 years were not included in the current study. For patients with more than one fracture in any of the inclusion periods, the first fracture was included in the present study.

Registration cards from the incidence studies of 1978/79 and 1988/89 were retrieved, and the cases were identified through name and date of birth. The data from the registration cards were transferred into an electronic register, and linked to the National Population Register (Norwegian Tax Administration, Oslo) to achieve the full unique Norwegian 11-digit identification number for each patient. The data from 1996/97 were stored in an electronic register with the 11-digit identification number.

In the 1988/89-cohort, 26 patients who had sustained a hip fracture in the previous inclusion period and 42 patients aged <50 years were not included. In addition, ten patients were not included due to: I) unidentifiable date of fracture (n=2); II) loss to follow up (n=4); III) missing unique identification number (n=3); IV) resident outside Oslo (n=1).

In the 1996/97 cohort, 15 patients had sustained a hip fracture in one of the earlier inclusion periods, and 21 patients were aged <50 years. All the remaining patients were included.

Data from the 5180 cases included (Table 1) were linked to The Cause of Death Register provided by Statistics Norway (Statistics Norway, Kongsvinger, Norway). The patients were followed up with respect to death from all causes until 31st of December 2007 (Additional file 1: Flow chart).

The study was designed as a historic cohort study. Exposure was sustaining a hip fracture in the registration periods. Outcome was death from all causes. Covariates were sex, age, and type of fracture.

The hip fractures were defined as femoral neck or intertrochanteric. Subtrochanteric fractures were not included.

Hip fractures in patients aged ≥65 years are often referred to as geriatric hip fractures, and younger patient are frequently not included in survival studies [8, 13]. Those aged ≥ 85 years have a high absolute mortality during the first six months of follow up, and a shorter duration of excess mortality than younger hip fracture patients [5, 7, 9]. The patients were therefore divided into following age groups: I) 50–64 years; II) 65–84 years; and III) ≥ 85 years.

In the 1978/79-cohort, fracture time was only registered by month and year. Date of fracture was consequently set to the first of the month to avoid negative survival time. To assess the influence of this fictive fracture date, survival analyses with the fracture date set to both the 1st and the 15th of the month were performed. The different fictive fracture days did not influence the results presented.

Comparisons of baseline data between the three cohorts were made using the Kruskal Wallis test and the One-Way-ANOVA. Kaplan-Meier-curves were estimated for each cohort stratified by sex, age group, and fracture type.

Age- and sex-specific one year-mortality rates for Oslo from 1978–2007 were provided by Statistics Norway and were used to calculate the expected survival curves for each cohort (Figure 1) as described by Therneau [19].

Standardized mortality ratio (SMR) expresses the level of excess mortality [20], and was calculated as the ratio of patient mortality to mortality in the background population of Oslo. The background population corresponded to the study population with respect to time period, sex, and year of birth. SMRs were calculated for the three cohorts for each sex, age-group, and fracture type, for the 0–6 months, 6–12 months, 0–1 year, 1–5 years and 5–10 years intervals after fracture. Sex- and age stratified analyses were also performed for the 0–1 year interval to allow comparison with earlier studies. Confidence intervals for SMR were computed as bootstrap BC_a intervals with 10 000 replications [21].

To assess the duration of the excess mortality, time-framed Kaplan-Meier curves for 5-years intervals starting at each year of the follow up time were calculated for each sex and age-group. These curves were compared with the corresponding expected survival curves. Only patients still alive at the start of the time interval were included. One sample log rank tests were used to test for statistical significance between observed and expected curves. The beginning of the last 5 year interval, where there still was statistical significance between the expected and observed curves, was set as duration of excess mortality.

The level of statistical significance was set at p<0.05. Analyses were performed using SPSS 14.0 (SPSS Inc., Chicago, IL, USA) and R 2.10.1 (The R Foundation for Statistical Computing, Vienna, Austria) with the R packages survival and boot.

The study is performed in compliance with the Helsinki Declaration. The Regional Committee for Research Ethics approved the study. A dispensation from professional secrecy was given by the Norwegian Directorate for Health and Social Affairs. Permission to handle sensitive information were sought and received from the Data Protection Agency.

The proportion of men was 21-22% in all three cohorts (Table 1). Mean age at time of fracture increased by about 5 years in both men and women from 1978/79 to 1996/97. The proportion of patients aged ≥ 85 years increased from 19 to 38%. The proportion of intertrochanteric fractures was higher in the 1996/97-cohort than in the 1978/79-cohort (p=0.003).

Kaplan-Meier curves of the three cohorts showed a substantially higher mortality for both sexes, in all age groups, than did the expected survival curves (Figure 1).

One-year mortality in the different cohorts ranged from 30% to 33% in men, and from 21% to 25% in women (data not shown).

The highest SMR was observed within 6 months after the fracture (Tables 2, 3, 4).

Men had higher SMR than women during all follow up intervals and in all three cohorts.

The SMRs were higher in those aged 65–84 years than in those aged ≥85 years during all follow up intervals and in all three cohorts (Table 3). There were no substantial differences in SMRs of femoral neck and intertrochanteric fractures within the cohorts (Table 4). Sex stratified analyses for each age group demonstrated higher SMR in men than in women in patients aged 56–84 and ≥ 85 years (Table 5). The lowest SMR was seen in the oldest age group in both sexes.

Changes in SMRs over the decades were more evident during the 0–6 month interval, than during long term follow up (Tables 2, 3, 4).

In women, there was a reduction in SMR from 1978/79 to 1996/97 during the 0–6 month interval. In men, a similar trend was observed (Table 2). In the age stratified analyses a statistically significant reduction in SMR from 1978/79 to 1996/97 was evident in the 0–6 months interval in those aged ≥85 years (Table 3). Further age- and sex stratified analyses performed for the 0–1 year interval revealed a statistically significant reduction in SMR only in women aged ≥85 years and a trend towards declining SMR in men aged 65–84 years (Table 5).

The changes in SMR from 1978/79 to 1996/97 were similar for femoral neck and intertrochanteric fractures (Table 4).

The longest duration of excess mortality was found in women aged 65–84 years (Table 6), where the excess mortality lasted more than 19 years in the 1978/79-cohort, for 14 years in the 1988/89-cohort, and until end of follow up in the 1996/97-cohort. The shortest period with excess mortality was seen in men aged ≥85 years. In this age-group, the mortality returned to the level of the background population after two to four years. Except for women aged ≥85 years, there was a trend towards shorter duration of excess mortality from 1978/79 to 1996/97.

The observed 1-year SMRs in the present study tended to be higher than in other similar reports. One study on hip fracture caused by falls from standing height or less in patients aged > 60 years showed a 1-year SMR of 2.18 for women and 3.17 for men [22]. Another study only including cognitively intact, ambulatory patients aged ≥65 years reported an overall 1-year SMR of 1.5 [8]. The present study includes practically all hip fracture patients aged >50 years in Oslo, including those living in nursing homes. Holvik et al. reported a one-year mortality of 46% in patients admitted from nursing homes, compared with 13.7% in patients living at home at the time of fracture [23]. In the present study, place of living is not registered in all three cohorts. However, in the 1988–89 cohort, 24% of the patients stayed in nursing home at the time of the fracture. Thus, the higher mortality among nursing home residents may have contributed to the high SMR in the current study. By also including those aged 50–65 years, average SMR is expected to increase because of the high relative excess mortality in this age group.

The higher relative excess mortality among the youngest compared with the oldest is in accordance with other reports [5, 7, 24]. The high impact of hip fracture on relative excess mortality among the youngest is probably related to the low mortality in the corresponding population, and the higher frequency of comorbidity in young hip fracture patients compared with the background population with same age [6, 18, 25, 26].

As in other reports, we found a higher excess mortality in men compared with women [5, 8, 10, 27]. Higher excess mortality in men is also reported in studies adjusting for the higher rates of complications and comorbidity [6, 10]. The reasons for the higher excess mortality in men remain unclear.

The highest excess mortality occurred within the first six months after the fracture, in accordance with earlier reports [5, 6, 11]. The high mortality in this period is probably a combined effect of the trauma and comorbidity [6, 11]. The lower SMR during long-term follow up is to be expected, as the impact of trauma declines with time.

Earlier reports on changes in mortality over the decades have been conflicting. A population-based British study showed a reduction in one-year mortality from 1968 to 1983, but no reduction from 1983 to 1998 [13]. Furthermore, a Danish register study reported a slightly increased excess mortality in hip fracture patients in the period 1986 to 2001 compared with 1981 to 1985 [9]. The methods used in these studies differ from the present one, which limit the possibility for comparison of the results.

The present data showed a statistically significant reduction in the 0–6 month SMR from 1978/79 to 1996/97 in patients aged ≥ 85 years. Decreased mortality following pneumonia and myocardial infarction [28‐31], which are common concomitant diseases in old hip fracture patients [12], may have contributed to the improved survival in the oldest age group. One may speculate that an increasing disease burden in hip fracture patients aged < 85 years, may be a reason for the stable SMR over time in these age groups. These speculations are supported by data from 2005 – 2009 from the Norwegian hip fracture register, which shows an increased proportion of patients with an ASA- (American Society of Anaesthesiologists) score [32] of 3 (severe systemic disease) at admittance, and fewer with an ASA score of 1 and 2 (none or mild systemic disease) [33]. It is not known when this trend started.

Although more pronounced in women than in men, there was a similar reduction in the excess mortality for both sexes during the 0–6 months interval from 1978/79 to 1996/97 in the analyses stratified on sex (Table 2). However, during the 6–12 months and the 1–5 years interval there was a trend towards increased SMR from 1978/79 to 1988/89 and a reduction from 1988/89 to 1996/97 in men, which was not found in women. The present data do not provide any explanation for this. Changes in incidence of, and survival after diseases more common in men than in women might have influenced the outcome in men [28, 34]. Furthermore, Bacon et al. found a greater improvement of survival in elderly men than in women in the period 1965 to 1993 [35]. This highlights the need for separate analyses of mortality in men and women.

The present data show that secular changes in SMR are most pronounced during short time follow up. This is to be expected, as the excess mortality is highest in the early phase after the fracture. Modifiable factors which could improve survival are time from fracture to operation [36], the use of prophylactic antibiotics [37], and early mobilisation [38].

Patients suffering from intertrochanteric fractures are supposed to be frailer and have more comorbidity and a higher mortality than those with femoral neck fractures [39], and the increased proportion of intertrochanteric fractures in our study could have influenced the results. However, in the present study, the SMRs were similar for the two fracture types, implying that fracture type is not a major determinant of excess mortality. The results are in concordance with other studies that have shown that the age- adjusted mortality in patients with intertrochanteric fractures is the same as for those with femoral neck fractures [40, 41].

Duration of the period with excess mortality ranged from two years to more than 19 years in the different sex- and age-groups. Hip fracture patients aged 65–84 years had, in accordance with other population based studies [5, 11], the longest duration of excess mortality. This prolonged period of excess mortality may reflect the high rate of comorbid conditions also among those surviving the first months after the fracture. This is further emphasised in other studies reporting prolonged excess mortality only in patients with a high degree of comorbidity [6, 11]. It is to expect that the oldest patients have the shortest period of excess mortality, as a high proportion in this age group will die during the follow up. However, the present study demonstrated an increase from 3 to 8 years of the period with excess mortality among the oldest women. Expected remaining life-time in Norwegian women aged 85 years was 5.9 years in 1996 according to the data provided from Statistics Norway, and changes in duration of excess mortality are therefore considered to be of clinical relevance. The increased duration of excess mortality may be a result of the reduced 6-months and 1-year excess mortality observed in hip fracture patients ≥85 years, which may have left a higher number of frail and sick patients for the long term follow up, contributing to excess mortality for a longer time period. Contrary, the shorter duration of excess mortality in men compared with women, may reflect the higher early mortality in men, leaving few and relatively healthy subjects for follow up.

Although, SMRs allow comparison of mortality in cohorts from populations with different background mortality, it may be considered as a limitation that no data on comorbidity or medication were available. Hip fracture patients are likely to have a higher frequency of comorbidity than the general population [6, 11, 18, 25], and changes in disease burden and medication in the fracture population may have differed from that in the background population. Another limitation of the study is that data were collected for only one calendar year in 1996/97, influencing the statistical power, particularly in men and the younger patients.

The strengths of the present study are the completeness of registrations of deaths and the long follow up of hip fracture patients from validated population-based incidence studies over three decades. No major changes in diagnostics took place in the actual period. The definitions remained the same, and the data collection was equal in the three incidence studies [14‐16]. No electronic diagnosis registers were available in 1978/79. However, the additional identification of the hip fractures through medical records and x-rays in all three incidence studies makes the data collection accurate and comparable in all three studies [14].