Project context

In 2008, orthogeriatric care at Oslo University Hospital was reorganized and became a part of the acute geriatric ward. The new service had the capacity to serve approximately half of the patients admitted with hip fracture. The remaining patients were treated in the orthopedic ward. To evaluate the new model, we randomly allocated patients between the acute geriatric and the orthopedic wards. The first hip fracture patient was admitted to the acute geriatric ward in June 2008 and after a pilot period inclusion in the study started in September 2009. The recruitment ended in January 2012. The study protocol containing further information is published elsewhere [17].

Study design

We carried out a randomized, controlled, single-blind trial comparing pre- and postoperative orthogeriatric care integrated in the acute geriatric ward to usual care in the orthopedic ward. Inclusion and randomization took place in the emergency department, overseen by the duty orthopedic surgeon. Allocation was by sealed, opaque, numbered envelopes. Randomization was based on computer-generated random numbers (blocks of variable and unknown size) and was carried out by a statistician (ES) not involved in the clinical service. Randomization was stratified according to whether or not the patients were admitted from nursing homes. Included patients were transferred directly from the emergency department to the allocated ward, and had their entire hospital stay in the same ward except for time in the operating theater and a few hours in the postoperative care unit. Operative and anesthetic procedures were the same in the two groups.

Study participants

All patients admitted acutely to Oslo University Hospital with a hip fracture (a femoral neck fracture, a trochanteric or a sub-trochanteric fracture) were eligible for inclusion. Patients were excluded if the hip fracture was a part of a high energy trauma (defined as a fall from higher than one meter) or if they were moribund on admission.

Intervention and control

A key element of the intervention was a Comprehensive Geriatric Assessment (CGA) as a basis for treatment planning. All team members (geriatrician, nurse, physiotherapist and occupational therapist) were expected to assess patients during their first day on the ward, and the team had daily meetings to co-ordinate treatment and to plan discharge. Clinical routines were developed based on a literature search, experience from earlier orthogeriatric models and the pilot phase prior to the start of randomization. Checklists were printed out and made immediately available for the treatment team for each patient. Details about the clinical routines have been published [17] and included medication reviews, early and intensive mobilization, optimizing pre- and postoperative nutrition and early discharge planning.

The control group was treated in the orthopedic ward, a 52 bed ward admitting a range of elective and non-elective orthopedic patients. The staff-patient ratio was similar to that of the acute geriatric ward (Table 1). There were, however, no multidisciplinary meetings and no geriatric assessments. Early mobilization was emphasized, and hip fracture patients were seen by a physiotherapist soon after surgery. The postoperative care unit was within the orthopedic ward, where all patients (including those allocated to intervention) were observed after surgery.

All patients included in the trial were offered a control in the orthopedic outpatient clinic four months after surgery. There was no additional intervention after discharge from hospital.

Measurements

Social and demographic information was collected during the acute stay. Information regarding surgical and anesthetic procedures, medical diagnoses (Charlson comorbidity index [18]), drug use and complications was also collected. Proxies were interviewed regarding pre-fracture Activities of Daily Living (Barthel ADL Index (BADL [19]) and Nottingham Extended ADL Index (NEADL [20])) and cognitive function (Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE [21])). Estimated height was derived using knee-heel length [22] and the patients were weighed using a chair scale. Mobilization after surgery was used as a process measure, recorded on day two post-surgery from case notes and observations. From September 2011, mobility was recorded with the activPAL™ body-worn sensor system [23]. The sensor was attached on the anterior aspect of the non-affected thigh as soon as possible after surgery and worn until discharge.

All patients were screened once daily for delirium using the Confusion Assessment Method (CAM) [24] preoperatively and until the fifth postoperative day (all) or until discharge (delirious patients). The study geriatrician or a study nurse completed all the assessments. If the nurse was unsure about the diagnosis, the study geriatrician was consulted. The CAM score was based on information from nurses, close relatives and hospital records related to the preceding 24 hours, in combination with a 10 to 30 minute interview with the patient. Tests of cognition, attention and alertness included the digit span test (forward and backward), orientation and delayed recall (from the Memorial Delirium Assessment Scale (MDAS) [25]). Delirium severity was measured with MDAS. Patients were assessed regularly on weekdays, but staff members who had been working during weekends were interviewed every Monday, and the case notes scrutinized in order to ascertain potential episodes of delirium. The mean number of delirium assessments during the stay was 5.7 (SD 2.7).

Follow up visits were carried out four and twelve months after surgery (with a time window of ± three weeks) by study nurses blind to allocation and to all clinical data during the original hospital stay. The patients were assessed in their current place of residence. Each visit typically lasted for two to three hours, and the evaluators started the assessment with the cognitive tests of the primary outcome.

At each follow-up visit, proxies were interviewed regarding physical (ADL) and cognitive function, using the same scoring systems as during the index stay. Mobility at the follow-up visits was assessed with the short physical performance battery (SPPB) [26]. Weight at follow-up was assessed using a standing scale that was calibrated to the chair scale used during hospital stay. Patients and proxies were asked about any hospital readmissions since surgery.

One specialist in geriatric medicine (TBW) and one specialist in old age psychiatry (KE) independently assessed whether the patients fulfilled the International Classification of Diseases, version 10 (ICD-10) criteria for dementia at baseline and 12 months after surgery. The assessors had access to all clinical data, but were blinded to allocation and delirium status during hospital stay. The inter-rater agreement upon the dementia diagnosis was satisfactory (kappa 0.87 at baseline and 0.83 at 12 months); disagreements were resolved through discussion.

Primary outcome

To construct the combined outcome measure, we normalized these scales into a 0 to 100 scoring (CDR had to be reversed since it is scaled in the opposite direction). The CDR carried a 50% weighting, and the immediate and delayed recall parts of the 10 word test each contributed 25% in the combined measure. Thus, a higher score on the primary outcome indicated better cognitive performance.

Secondary outcomes

Secondary outcomes included preoperative delirium, delirium severity, length of stay, mortality, mobility, place of residence, ADL function and weight changes at the follow up controls. CDR and the 10 words test were analyzed separately, in addition to other measures of cognition (Mini-mental state examination (MMSE) [31], clock drawing test [32], IQCODE).

Statistical analyses

No pre-trial data were available to carry out precise power estimates. Based upon previous experience with the CDR, we judged 300 patients to be sufficient to detect clinically meaningful differences [30]. As 20% of hip fracture patients can be expected to die within four months of surgery, we aimed to randomize 370 patients. Recruitment ended after randomization of 332 patients due to resource constraints.

A statistical analysis plan (SAP) was developed (and published online) prior to un-blinding of the data [33]. The primary analysis was carried out blind to allocation by the study statistician (ES).

The primary analysis was carried out as a modified intention-to-treat analysis including patients with CDR and a complete 10-word test at the four-month control. Two patients were sent to the ward opposite to randomization allocation, and these patients were analyzed according to the group in which they were treated (Figure 1). Three moribund patients (two randomized to the acute geriatric ward and one to the orthopedic ward) were recruited in error, and were excluded from the primary analysis.

The primary outcome was not normally distributed but the sample size was large and parametric methods could therefore be applied. To adjust for any inequality in the distribution of important prognostic variables between the intervention and control group, we performed a linear regression with the primary outcome as the dependent variable, and variables with known or believed influence on the outcome were included in the model in a stepwise manner, in addition to the randomization group. If their introduction to the model changed the effect estimate for the randomization variable by 10% or more, they were included in the final model. Variables were removed by stepwise backwards elimination until the final model was reached. Age (negatively skewed) and waiting time to surgery (positively skewed) had non-normal distributions, and were squared and log transformed, respectively, to achieve better fit of the model. Secondary outcomes were analyzed by the Mann–Whitney test, t-tests and Chi-square tests depending on data distribution. Pre-planned subgroup analyses were carried out in patients admitted from nursing homes, and in patients with and without pre-fracture dementia.

All statistical analyses were performed using IBM SPSS Statistics version 20, except for median differences and corresponding 95% confidence intervals that were estimated by the Hodges Lehmann estimator using StatXact 8.0.

Sensitivity analyses

Ethical considerations

The study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from the patients or substitute decision-makers if patients did not have capacity to consent. The study was approved by the Regional Committee for Ethics in Medical Research in Norway (REK S-09169a) and the Data Protection Officer at Oslo University Hospital (Ref. 1361).

Impact of intervention during hospital stay

The median length of stay was three days longer in the intervention group (median eleven versus eight days, P ≤0.001). Patients in the intervention group had a longer waiting time for surgery, but this difference was not statistically significant (median 26 versus 24 hours, P = 0.54).

There was a trend to greater mobilization in the intervention group on the second day after surgery (86% versus 80%). In 46 patients, mobilization after surgery was assessed with activPAL™ activity sensors. During the first five days after surgery, the patients were mobilized for a longer time in the standing or stepping position in the intervention group (median 29 minutes versus 17 minutes).

Primary outcome - cognitive function four months after surgery

Secondary outcomes and subgroup analyses

Patients treated in the acute geriatric ward performed better on all cognitive measures; CERAD immediate recall (median 12.5 versus 11.5) and delayed recall (median 3 versus 2) , approved clock drawing test (49% versus 40%), MMSE (median 24 versus 23) and CDR (1.5 versus 2.5). They also had better ADL function measured with the BADL Index (median 17 versus 16) and NEADL (median 26.5 versus 22). None of these differences were, however, statistically significant.

Patients randomized to the acute geriatric ward had better mobility four months after surgery, measured with SPPB (median 4 versus 3, 95% confidence interval for the median difference 0 to 2; P = 0.13) (Table 4). This difference was statistically significant in the pre-specified subgroup analysis restricted to patients living in their own home before the fracture (median 6 versus 4, 95% confidence interval for the median difference 0 to 2; P = 0.04). Subgroup analyses stratified according to pre-fracture dementia status and nursing home residence gave no other significant differences, except that patients from nursing homes randomized to intervention were more often mobilized the second day after surgery (Additional files 1, 2 and 3).

Twenty-eight (17%) patients treated in the acute geriatric ward and 24 (15%) treated in the orthopedic ward were dead four months after the surgery (P = 0.50). In both groups, 21% of the patients were readmitted during the first four months after surgery. The results at the 12-month follow up were similar to those after four months.

Sensitivity analyses

Several sensitivity analyses were performed and they showed no substantial differences from the primary analysis.

Strength and weaknesses

The main strength of this study was the randomized controlled design with blinded outcome assessments. Also, the inclusion of process measures, such as objective mobilization scores, confirms that the intervention was being delivered as intended. The inclusion of patients from nursing homes and those with dementia enhances generalizability as such patients are frequently excluded from trials [34]. On the other hand, nursing home patients are so frail and cognitively impaired that they may be unlikely to benefit from the intervention. To assess the efficacy in such patients, other outcomes than those we chose might be more feasible [35]. The combined outcome measure was designed to measure cognition in patients representing a broad spectrum of cognitive function and was based upon well-validated components. However, the scale combination has not been validated and, thus, we cannot be sure that it was sensitive to the intervention. As with all service evaluations, blinding of assessments during hospital stay was impossible and may have introduced bias.

Inclusion was terminated before the intended sample size aim was reached. However, as there were few differences between the groups in any of the secondary cognitive outcomes, and sub-group analyses also failed to show any substantial differences, it is reasonable to conclude that this intervention had no effect on cognition.

Comparison with other studies

The impact of orthogeriatric intervention on long-term cognitive function has not previously been assessed. Several studies have demonstrated that orthogeriatric care can prevent delirium in hip fracture patients. A recently published non-randomized controlled trial from Belgium [15] showed that an intervention provided by an inpatient geriatric consultation team was effective in reducing the incidence of delirium (37.2% versus 53.2%, P = 0.04), in keeping with a similar American RCT [16]. In both the American and the Belgian studies, all patients received standard treatment from the orthopedic team, whereas in our model orthopedic treatment (besides surgery) was limited to consultation service. A possible explanation for the lack of effect of our model could, therefore, be limited access to orthopedic expertise.

Few studies have compared pre- and postoperative intervention provided in a geriatric ward with usual care in an orthopedic ward. In comparison with usual care, such models have shown promising results, but cognition has seldom been assessed [5, 6]. To our knowledge, the only RCT evaluating a geriatrician-led fracture service (were geriatricians have the primary responsibility for the patients) is a Swedish study [36]. Although no preoperative intervention was included, the study showed that significantly more patients allocated to intervention regained independence in personal ADL performance at four and twelve months after surgery. The model was also effective in preventing postoperative delirium and reducing delirium duration [14]. In spite of the fact that we also included preoperative intervention, we were not able to prevent delirium. A likely explanation is that usual care was better in our study since the delirium rates both in the intervention and the control group were lower than in the Swedish study. The orthopedic ward in our study provided a short waiting time for surgery, similar staffing as in the geriatric ward, personnel with earlier experience with orthogeriatric models and delirium prevention, physiotherapy for most hip fracture patients, and an integrated post-operative care unit.

Orthogeriatric intervention is often reported to reduce waiting time for surgery (see Liem [37] for an overview). In our study, however, the waiting time for surgery was two hours longer in the intervention group. Both the intervention (26 hours) and the control group (24 hours) waited, however, for a short time compared to other orthogeriatric studies reporting a waiting time of two to three days and even longer [38–42], indicating that the control group received a good quality service.

Mobility has been assessed in several studies, but mostly by questionnaire. Some, but not all, studies have found that orthogeriatric services provide better mobility [36, 40, 43, 44]. In our study there was an overall trend that patients treated in the intervention group performed better at SPPB four months after surgery, and the difference was statistically significant in those living in their own homes before surgery. A difference on SPPB of 0.5 is considered clinically meaningful, and the effect seen in our study (six versus four points) is likely to be important and should be further explored in future studies.

Interpretation of the results

Despite our comprehensive intervention, the effect on the primary outcome was limited. There are several possible explanations for this. First, the choice of cognitive function as the primary outcome may have been too ambitious. For the intervention to be effective in this regard, two pre-suppositions had to be true. First, the orthogeriatric intervention had to be effective in reducing delirium. However, our intervention failed to prevent delirium or reduce delirium severity. This might be explained by the good quality of usual care at the orthopedic ward in our hospital, combined with sub-optimal circumstances in an often over-crowded acute geriatric ward.

Secondly, the primary outcome assumes that delirium lies on the causal pathway towards the development of dementia. Since delirium usually occurs in relation to acute illness, it is challenging to design studies that can address this question, but some evidence exits suggesting that delirium is associated with long term cognitive decline [13, 45, 46]. Our study is in keeping with this; the regression analysis showed that delirium was associated with a poorer score on the primary outcome, also when adjusting for potential confounders.

The study may have influenced treatment in the control group. The patients in the orthopedic ward were assessed daily, and in order to make a precise delirium diagnosis personnel in the orthopedic ward were interviewed regarding the patients cognitive status. This inevitably raised the awareness of delirium in the orthopedic ward.