Relative survival following hemi-and total hip arthroplasty for hip fractures in Sweden

Since 2005 the SHAR registers hip fracture patients whom undergo both total and (or) hemiarthroplasty. We identified all patients operated between 2005 and 2012, and this patient group served as our study group. If a patient had sustained bilateral hip operations due to fractures, we only considered the first operation (Fig. 1). To adjust for confounding, accessible and relevant variables were collected from three registers. SHAR provided data about the patients’ age at operation, sex, prosthesis type (hemiarthroplasty or total hip arthroplasty), and hospital type (rural, county, university or private). The SHAR is a Swedish quality register and has a high completeness (98%) and full coverage (100%) [8]. Date of death is continuously updated in SHAR. The risk of leakage out of the register was considered, however only a handful of patients are known to have asked for their data to be removed [9]. Emigration rate in this group of frail elderly Swedes can be considered very low, and similar to the general population of the same age. Those few patients who emigrated were censored as they were dead [9]. Using the Swedish unique 10 digit personal identity number we linked the data from SHAR to Statistics Sweden and the National Patient Register (NPR) (National Board of Health and Welfare) [10]. Statistics Sweden provided data about the patients’ education and civil status, while the NPR delivered data about the patients’ health status encoded as ICD-10 codes. Swedish hospitals are obliged by law to report all primary and secondary ICD-10-codes to the NPR for every contact with the hospital. Only individual with entries in all three registers were included. Individual ICD-10-codes were combined into Elixhauser comorbidity indices (ECI) for each patient [11, 12]. The ECI is an unweighted comorbidity index representing the number of present comorbidities out of 31 possible. The ICD-10-codes were identified from hospital based healthcare in the year before the index hospital admission. Patents without recorded visits were classified as patients without known comorbidity (i.e. Elixhauser = 0).

We summarized continuous variables as means and standard deviations, categorical variables as counts and percentages. We used Student’s t-test and χ2-test for group comparisons. In this paper we used relative survival analysis [13, 14] The measure of interest was cumulative relative survival function r(t). We estimated the cumulative relative survival function as \( \mathrm{r}\left(\mathrm{t}\right)=\frac{{\mathrm{S}}_{\mathrm{O}}\left(\mathrm{t}\right)}{{\mathrm{S}}_{\mathrm{P}}\left(\mathrm{t}\right)} \).

Where S_O(t) denotes the observed survival and S_P(t) the population or expected survival at time t. We extracted the Swedish population or expected survival S_P(t) from publicly available mortality tables tabulated for sex and age maintained by the Human Life-Table Database [15] and Human Mortality Data Base [16].

The cumulative relative survival function r(t) can be any non-negative number. If r(t) = 1, then the observed survival of the studied group coincides with the expected or population survival. If r(t) > 1 then the observed survival of the studied group exceed the expected or population survival. If r(t) < 1 then the studied group experiences an excess mortality. We can partition the observed hazard for each individual λ_O(t) into two additive components: λ_O(t) = λ_P(t) + λ_E(t).

The age- and sex-specific baseline hazard is denoted by λ_P(t). The difference between the baseline hazard and observed hazard, λ_E(t) denotes the excess hazard of the study group.

Alongside with the cumulative relative survival, the excess hazard and the cumulative incidence curve served as the main methods of analysis. Additionally, we attempted to model the effect of clinical and patient related covariates with the help of multivariable Cox Proportional Hazards Model in Transformed Time [17]. We tested the assumption of proportionality with graphical examination and Brownian bridges [18]. We observed violation of the proportionality assumption for prosthesis type, age and ECI. We modelled these variables by the introduction of a step function that split the data in two epochs up to 14 months (420 days) and the subsequent period until end of study period. The multivariable regression analysis included interaction term between the aforementioned variables and the step function for time.

The mortality rate of males was higher, and both sexes had survival rates inferior than their peers from the general population (Fig. 3). Younger age, analysed as a continuous variable, led to an increased relative risk of dying, being more pronounced in the later period.

In general, the higher the ECI the lower the survival was. But the group with a ECI of 1 had better survival during the first 420 days, compared with those with ECI of 0 (Fig. 3). Over time the negative effect of the increasing ECI on survival became more obvious.

Patients with total hip arthroplasty had a better survival than patients with hemi-arthroplasty. Of the hospital types considered university hospitals had lower survival rate during the first 6 years, whilst patients operated at private, rural and county hospitals had similar survival rates. Over time, the impact of implant-type decreased (Fig. 3).

The separation of the survival rates for the different socioeconomic status was significant for education. Patients who completed higher levels of education had better survival that patients with low or middle education levels. (See Fig. 4) The later two categories had similar survival rates during the whole follow up period. Being married compared to being single and having higher achieved levels of education lowered the risk of death, however being a widow did not (Table 2). Multivariable relative survival regression corroborated most of the results of the univariate analysis.

Swedish hip fracture patients whom undergo hip arthroplasty had a high excess hazard of dying, up to 100 days after the operation compared with individuals in the general population of the same age and sex. This excess hazard diminished with time but never subsided to negligible levels. The mortality rate of hip fracture patients remained higher than the mortality rate of the general population during the whole 8 years’ follow-up period.

Earlier papers have related the long-term mortality among hip fracture patients to the mortality rates in the general population [3, 4, 6, 19]. These studies highlighted an increased mortality among hip fracture patients compared with the general population using age and sex standardization. We used relative survival rate and multivariable modelling and could therefore relate every patient to the survival rate of members of the general population of the same age and sex.

The excess early mortality can possibly be explained by the comorbidities and frailty of hip fracture patients, aggravated by the injury and the surgery. The relationship between fracture and increased risk of death up to 8 years post-surgery, is more difficult to disentangle. Apart from the chronic disease status in this patient population, previous research has suggested that pain and fear of falling [20] is contributing to a decrease in activity and exercise and a decrease in walking distance [21], which in turn potentially contributes to a loss of independence and depression [22]. In order to improve outcomes following hip fracture surgery, it is imperative to identify modifiable factors that could possibly aid survival and function [23]. There have been promising results where an improved and close cooperation between orthopaedic surgeons and geriatricians have led to a reduction in mortality [24] aided by systematic and sustained quality improvement efforts [25, 26].

Advanced age is associated with an increase in mortality after hip fracture however, this may be due to the higher general mortality rates that comes with aging [27, 28], as we by relative survival analysis observed that age was positively associated with the relative survival. Patients who suffer a hip fracture in their younger years lost more of their expected life span than patients who had their fracture in more advanced age. This effect increased with the follow-up time. This finding suggests that especially in the younger hip fracture population, there could be a greater benefit of an improved collaboration between healthcare professionals within the hospital setting, general practitioners and the municipal social care.

The excess mortality of males was higher than for females. This was more evident in the beginning of the study period. Subsequently the excess sex-related differences in survival following hip fracture surgery follow a similar trends of the sex-related difference in survival of the general population. That men have a higher risk of dying after hip fracture is well known from the literature, and was shown also in an recent study [19]. It can be debated if this is induced by sex as an isolated variable, or if being male is a surrogate variable for comorbidity, unhealthy lifestyle or other confounders adding to their risk of both hip fracture and death.

Generally patient comorbidity had a significant adverse effect on survival [28], even though moderate comorbidity may not be associated with an excess one-year-mortality [27‐31]. The hazard rates increased with increased score of comorbidity, as measured by the ECI [32]. This effect became stronger with time. Interestingly, in the early period patients with ECI = 1 have higher survival rate than patients with ECI = 0, while after 420 days no such difference was seen. An attempted explanation could be that patients with known comorbidities are receiving increased medical attention prior to the surgery, improving their health status. The group of patients with an ECI = 0 may consist of two types of patients; really healthy individuals and those who are not, but never sought medical attention and as such have undiagnosed and/or untreated comorbidity. Assumingly, in the hip fracture population there are a number of patients neglecting their health. These patients may present with an unstable medical condition at the time of their hip fracture, and subsequently have a higher risk of dying. We believe our results could question the validity of comorbidity indices based on administrative patient registers. Besides sick individuals not being known to the healthcare, there is a possibility of misdiagnosis bias due to incorrect coding [33].

We found lower relative survival of patients treated at university hospitals. Similar results has been described in a Danish study [34], whilst a Canadian study found an improved survival in teaching hospitals compared to community hospitals [35]. Finally, a study from the USA concluded that hospital volume did not predict mortality, even though hospitals with high caseload had fewer complications [36]. Swedish university hospitals typically serve an urban population, so when comparing different hospitals, we also compare different populations. Based on the available evidence one could conclude that both hospital level and volume are blunt variables confounded by case-mix, level of competence of the staff and an unclear definition of what high volume is, making comparisons between different healthcare systems difficult.

The year of surgery was not associated with any excess risk of mortality, indicating that during our study period the rate between hip fracture patients survival and the survival of the general population did not change. This finding is in agreement with the findings of Klop et al. [37]. In a systematic review on randomized clinical trials, Mundi et al. found similar mortality in hip fracture patients over time [6].

Patients operated with total hip arthroplasty had lower mortality than those with hemi-arthroplasty. This is most probably a reflection of selection bias as choice of implant is based on patient frailty, age and activity level.

Solitary life style [38] and social deprivation [39] have been previously associated with increased mortality after hip fracture. In our study we found similar results, where being non-married brought on a slight increase in mortality. However, in our study being widowed had no adverse effect on mortality.

Other factors may influence mortality. Solbakken and collaborators [40] concluded that self-perceived health, smoking, and BMI predict mortality of hip fracture patients. Vosoughi and collaborators [41] reached similar conclusions regarding smoking and BMI and noted improved nutrition as modifiable factors. Hence, as our register data is limited in terms of such variables, our results could not be adjusted for those and might have to be interpreted with caution due to varying degree of validity [42].

The SHAR has an excellent completeness of over 98% [8], in combination with the large patient cohort are the main strengths of the present study. The additional access to socioeconomic and comorbidity data from governmental administrative registers further add to the strength [10]. Additionally, the use of relative survival analysis has an overarching advantage that, through a single measure, it indicates the degree to which a study sample corresponds to the general population with respect to health status and survival [43].

Orthopaedic healthcare of today focuses on a timely and effective treatment of hip fracture patients. Post-operative care can also be of a fast-track nature, both reflecting new scientific knowledge and health economic challenges. Concerns have been raised whether a shorter length-of-stay is associated with an increase in mortality [44]. Our findings may assist the clinician to identify patients at risk of excess mortality. The patient without prior medical records or diagnoses can actually be more challenging than expected. A younger patient will suffer a risk of a greater expected life span loss than the more elderly patient and this should not be neglected. The same goes for singles and/or males. The hip fracture patient in general carries a highly increased risk of dying during several months after fracture, compared to the general population, and further closer follow-up care have to be provided after discharge from the hospital.