Abstract

Background

The Nottingham Hip Fracture Score (NHFS) was developed and validated in a single centre in 2007 as a predictor of 30 day mortality. It has subsequently been shown to predict longer term and functional outcomes. We wished to assess the ability of NHFS to predict outcomes in other centres and to investigate the change in outcome after hip fracture over time.

Methods

The NHFS was calculated for all patients with data from three UK hip fracture units: Peterborough (1992–2009), Brighton (2008–9), and Nottingham (2000–9) including 4804, 585, and 1901 patients, respectively. The logistic regression was used to recalibrate the NHFS to 30 day mortality across the three units using a random selection of 50% of the data set. Calibration was assessed using the Hosmer–Lemeshow goodness of fit.

Results

The median (inter-quartile range) NHFS values were Peterborough [4.0 (1–6)], Brighton [5.0 (3–7)], and Nottingham [5.0 (3–7)]. There was no correlation between 30 day mortality and time (R2=0.05, P=0.115). The proportion of patients with NHFS≥4 showed a weak correlation with time (R2=0.2, P=0.003). The original NHFS equation overestimates mortality in the higher-risk groups. A modified equation shows good calibration for all three centres {30 day mortality (%)=100/1+e[(5.012×(NHFS×0.481)]}. The hospital was not a predictor of 30 day mortality.

Conclusions

The NHFS, with an updated equation, is a robust predictor of 30 day mortality after hip fracture repair in geographically distinct UK centres.

Editor's key points

  • The Nottingham Hip Fracture Score (NHFS) has been developed to predict outcome after hip fracture surgery.

  • The performance of the NHFS was studied in three large UK hip fracture units including 7290 patients.

  • Recalibration of the original equation resulted in a robust tool for predicting 30 day mortality after hip fracture applicable to multiple UK centres.

Hip fracture in the elderly is associated with a high postoperative mortality, poor functional outcome, and significant financial and social costs.1,2 The accurate prediction of risk is important for informed decision-making, communication with patients and relatives, planning of care, and internal or external audit of outcomes. The Nottingham Hip Fracture Score (NHFS) was originally developed and validated for a single-centre data set from 2000 to 2007.3 It has been shown to be reliable in predicting 30 day mortality,3 1 yr mortality,4 and functional outcome,5 and has been used for stratification in a clinical trial.6 Although it has been promoted as a tool for use in other units,7 to date there have been no published data on its performance elsewhere. There has been a downward trend both in the prevalence of hip fracture1 and in postoperative mortality in the UK in recent years,8 and the original predicted mortality might no longer be valid due to improvements in multi-disciplinary care. We therefore wished to investigate the performance of the NHFS across separate large hip fracture units.

Methods

Databases

The Nottingham hip fracture database has been described previously.3 It contains preoperative social, functional, medical, and physiological data; surgical, anaesthetic, and orthogeriatric interventions; and outcome data including length of stay, complications, discharge destination, and mobility at discharge. These data are collected prospectively by dedicated audit officers; internal audit has demonstrated high levels of accuracy. The Peterborough (1992–2012) and Brighton9 (2005–2012) databases are similar, although with slightly different data captured. All three data sets allow for the calculation of the NHFS and determination of 30 day postoperative mortality. Data used in the original derivation of NHFS from Nottingham were not included in the calculation of scores.

Nottingham Hip Fracture Score

The NHFS is a summative score of seven preoperative variables which give an estimated risk of 30 day postoperative mortality using a standard logistic regression (Tables 1 and 2).

Table 1

Nottingham Hip Fracture Score. MMTS, mini-mental test score. Original predicted 30 day mortality is calculated by substituting the total NHFS into the equation: 30 day mortality (%)=100/1+e[4.7182−(NHFS/2)]. New predicted 30 day mortality is calculated by substituting the total NHFS into the equation: 30 day mortality (%)=100/1+e[5.012−(NHFS×0.481)]

Variable Value Score Proportion in complete data set (%; n = 11 670) 
Age <66 3.8 
 66–85 yr 58.3 
 ≥86 yr 36.9 
Sex Male 21 
Admission Hb ≤10 g dl−1 10 
MMTS ≤6 out of 10 33 
Living in an institution Yes 27 
Number of co-morbidities ≥2 25 
Malignancy Yes 
Variable Value Score Proportion in complete data set (%; n = 11 670) 
Age <66 3.8 
 66–85 yr 58.3 
 ≥86 yr 36.9 
Sex Male 21 
Admission Hb ≤10 g dl−1 10 
MMTS ≤6 out of 10 33 
Living in an institution Yes 27 
Number of co-morbidities ≥2 25 
Malignancy Yes 
Table 2

Predicted values for 30 day mortality by the NHFS

Total NHFS Predicted 30 day mortality (%)
 
 Original NHFS New NHFS 
0.9 0.7 
1.5 1.1 
2.4 1.7 
3.8 2.7 
6.2 4.4 
9.8 6.9 
15 11 
23 16 
33 24 
47 34 
10 57 45 
Total NHFS Predicted 30 day mortality (%)
 
 Original NHFS New NHFS 
0.9 0.7 
1.5 1.1 
2.4 1.7 
3.8 2.7 
6.2 4.4 
9.8 6.9 
15 11 
23 16 
33 24 
47 34 
10 57 45 

The NHFS was calculated for all patients aged more than 60 with complete data for each of the units and predicted and observed 30 day mortality calculated (Table 3).

Table 3

Numbers of patients according to site and NHFS in the development sets

NHFS Peterborough
 
Brighton
 
Nottingham
 
Overall
 
Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of total) 30 day mortality 
42 0.0 13 0.0 59 0.0 
31 3.2 14 0.0 47 2.3 
21 0.0 32 
529 22 1.7 41 14 173 19 1.7 743 21 1.6 
652 27 5.1 77 26 1.3 261 28 3.1 990 27 4.3 
575 24 8.9 89 30 9.0 224 24 7.1 888 25 8.5 
375 16 12.0 42 14 9.5 175 19 9.7 592 16 11.2 
145 15.2 24 25 50 14 219 16.0 
20 35.0 12 16.7 14 46 19.6 
0.08 50.0 0.7 0.0 0.2 0.0 0.03 16.7 
Total 2393   296   934   3623   
NHFS Peterborough
 
Brighton
 
Nottingham
 
Overall
 
Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of total) 30 day mortality 
42 0.0 13 0.0 59 0.0 
31 3.2 14 0.0 47 2.3 
21 0.0 32 
529 22 1.7 41 14 173 19 1.7 743 21 1.6 
652 27 5.1 77 26 1.3 261 28 3.1 990 27 4.3 
575 24 8.9 89 30 9.0 224 24 7.1 888 25 8.5 
375 16 12.0 42 14 9.5 175 19 9.7 592 16 11.2 
145 15.2 24 25 50 14 219 16.0 
20 35.0 12 16.7 14 46 19.6 
0.08 50.0 0.7 0.0 0.2 0.0 0.03 16.7 
Total 2393   296   934   3623   

A revised equation for converting the NHFS to the predicted mortality was derived using the complete combined data for 1992–2009 and standard logistic regression with NHFS and hospital as the two predictors (backward stepwise logistic regression, using 0.05 for entry and 0.1 for removal). A training set of 50% of subjects, chosen at random, was used to generate the equation with the remaining 50% used for validation. The goodness-of-fit was assessed for training and validation sets with the Hosmer–Lemeshow tests. We specified that adequate goodness-of-fit would be achieved if P>0.05.

Results

A total of 7290 patients were available for analysis of 30 day mortality and the NHFS (Table 2). The overall 30 day mortality for the complete cohort was 6.6%. The mortality was related to the NHFS in the overall cohort and for each centre individually (Tables 3 and 4).

Table 4

Numbers of patients according to site and NHFS in the validation sets

NHFS Peterborough
 
Brighton
 
Nottingham
 
Overall
 
 Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of total) 30 day mortality 
40 0.0 0.4 18 0.0 59 0.0 
38 0.0 10 50 1.3 
0.3 0.0 0.0 0.0 14 0.0 
549 23 2.2 29 10 3.4 138 14 2.2 716 20 2.1 
643 27 3.7 84 29 3.6 307 32 3.6 1034 28 3.8 
591 25 8.0 85 29 2.1 254 27 8.7 930 25 8.6 
389 16 12.3 45 16 8.9 162 17 8.0 596 16 11.3 
134 13.4 32 11 15.6 62 11.3 228 12.3 
19 26.3 10 20.0 12.5 37 28.1 
0.04 0.0 0.4 100 0.10 0.08 0.0 
Total 2411   289   967   3667 0.08 0.0 
NHFS Peterborough
 
Brighton
 
Nottingham
 
Overall
 
 Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of unit total) 30 day mortality Number (% of total) 30 day mortality 
40 0.0 0.4 18 0.0 59 0.0 
38 0.0 10 50 1.3 
0.3 0.0 0.0 0.0 14 0.0 
549 23 2.2 29 10 3.4 138 14 2.2 716 20 2.1 
643 27 3.7 84 29 3.6 307 32 3.6 1034 28 3.8 
591 25 8.0 85 29 2.1 254 27 8.7 930 25 8.6 
389 16 12.3 45 16 8.9 162 17 8.0 596 16 11.3 
134 13.4 32 11 15.6 62 11.3 228 12.3 
19 26.3 10 20.0 12.5 37 28.1 
0.04 0.0 0.4 100 0.10 0.08 0.0 
Total 2411   289   967   3667 0.08 0.0 

There was a statistically significant difference between the NHFS for the three sites (Kruskal–Wallis, P<0.001). The median (inter-quartile range) was: Peterborough [4.0 (1–6)], Brighton [5.0 (3–7)], and Nottingham [5.0 (3–7)].

There was no correlation between 30 day mortality and time (R2=0.05, P=0.115) (Fig. 1). The proportion of patients with the NHFS≥4 showed a weak correlation with time (R2=0.2, P=0.003).

Fig 1

The longitudinal change in 30 day mortality after hip fracture repair in all three units studied. Thirty day mortality is shown for each quarter. There is no significant correlation between time and 30 day mortality (R2=0.027).

Fig 1

The longitudinal change in 30 day mortality after hip fracture repair in all three units studied. Thirty day mortality is shown for each quarter. There is no significant correlation between time and 30 day mortality (R2=0.027).

Overall, the original NHFS equation overestimates mortality in the higher-risk groups (Tables 2–4).

The hospital was not a significant factor in logistic regression using both the NHFS and site as predictors. The logistic regression using the NHFS as the sole predictor gives good calibration for all the development sets (by site and overall) and on the validation sets (all P>0.1).

The proportions of patients (from the complete data set, including the original cohort) with the individual score items are given in Table 1 (11 670 patients in total).

Discussion

These data suggest that the NHFS, a scoring system for assessing the risk of 30 day mortality after hip fracture repair, is applicable outside its original derivation cohort. The original equation overestimated risk at lower and higher scores. A revised equation, using the same relative attribution of risk factors, has good calibration across all three units in both development and validation sets. In particular, the score is accurate for the most prevalent group with NHFS 4–6.

The risk status of those with hip fracture, as measured by the NHFS, increased over time, suggesting that the population is becoming more frail. These data are intentionally from before the introduction of Best Practice Tariff.10 Nationally, there has been a downward trend in mortality over the last 2 yr that might not have been seen in these data.

We chose to recalculate the relationship between 30 day mortality and the original NHFS rather than create a new scoring system based on complete re-analysis of the data. We chose to do this for theoretical and practical reasons. First, the ordinal relationship between the NHFS and 30 day mortality is apparent in the data for the individual units and the combined data set supporting the premise that the NHFS is a predictor of outcome. Secondly, the individual components of NHFS are well-recognized independent predictors of outcome after hip fracture, so re-analysis was unlikely to produce major differences in predictor variables. Thirdly, the NHFS is already used in its current form, so changing the variables is, on a practical level, not helpful. The association between NHFS and 1 yr mortality and functional outcome remains valid as these studies stratified patients based on their absolute NHFS rather than on the predicted 30 day mortality. In line with previous studies, most notably POSSUM/P-POSSUM,11 we chose to treat the NHFS as a single predictor variable.

To have clinical credibility, any scoring system should ideally be demonstrated to be robust in more than its original derivation cohort. The original work demonstrated validity in a validation data set, but this was limited to Nottingham data. These new data demonstrate that the NHFS is a robust tool in clinically, demographically, and geographically distinct units within the UK. This would suggest that further multi-centre validation across the UK would be appropriate. Given the differences that exist in models of care and populations between the UK and other countries, we do not know whether the NHFS has validity in other countries. There is no perfect scoring system that combines the ease of use and relevant clinical validity. The NHFS is a compromise between simplicity (such as the ASA classification) and complexity. The calculated coefficients from the original derivation of the NHFS are not integer values and were deliberately rounded to create an accessible score at the expense of some accuracy.

Although calibration of NHFS with 30 day mortality in the complete set and for individual units is acceptable, the proportion of patients with a high (>4) NHFS appears to be increasing slightly with time. This would suggest that outcomes for these patients are improving, despite an increase in frailty. However, it does mean that in the future, the NHFS is likely to require recalibration to these improved outcomes.

These data, which represent a sample size equivalent to 17% of the annual number of hip fractures in England,1 also provide useful information regarding individual factors relevant to those caring for people with hip fracture. Marked cognitive dysfunction (AMTS≥6) on admission is common, occurring in around one-third of people. This frequency supports the new UK NICE hip fracture quality standard for hip fracture that all patients should be assessed for confusion on admission.12 It also has implications for research studies. We would argue that studies in hip fracture should have to provide strong justification of why confused patients should not be included. Severe anaemia (Hb<10 g dl−1), a known risk factor for poor outcome,13,14 is common affecting 10% of patients, and coupled with the blood loss in the perioperative period should be a strong driver towards research aimed at managing this effectively. Other workers have also demonstrated the association between perioperative anaemia and poor outcome in the elderly.15 The recent FOCUS study16 suggested no benefit of a liberal postoperative transfusion threshold in relation to mobility or mortality, although cardiac complications were higher in the restrictive group. Other workers have suggested that haematopoietic agents (i.v. iron, erythropoietin) confer benefit without the risks (and costs) of transfusion.17 Frailty is recognized as a marker of poor outcome in hospital and community settings. To an extent, the NHFS is a summary measure of various surrogates of frailty in the elderly. Future work might examine the clinical utility of extending the application of the NHFS to non-hip fracture areas albeit with the modification of outcome such as mortality after medical admission, laparotomy, or pubic ramus fracture.

Although randomized controlled trials are viewed as the gold standard for identifying effective treatment, high-quality databases allow for the observational description of ‘all’ patients undergoing a specific procedure and analysis of pragmatic interventions such as the type of anaesthesia18 in comparison with very much smaller randomized controlled trials of subgroups of the hip fracture population, extrapolated to the hip fracture population as a whole.19 National data collection would assist in this process: the National Hip Fracture Database collects most of the NHFS fields already. Prospective collection of the three data fields not currently collected by NHFD (Hb, Cancer, two or more co-morbidities) would confirm (or refute) the accuracy of the NHFS, and provide baseline score for future comparison, both between and within units and longitudinally over time.

On the basis of the data presented here, we believe that the NHFS is a robust tool that can be used by clinicians at the individual patient level to communicate risk with patients and carers, to help plan care, and at the unit level to assist with the audit of data.

Declaration of interest

I.K.M. has received honoraria from Schering-Plough in the past 5 yr. He is a member of the Editorial Board of the British Journal of Anaesthesia. He is a member of the NICE Quality Standards Topic Expert Group on Hip Fracture. M.P. has received expenses and honoraria from a number of commercial companies and organizations for giving lectures on different aspects of hip fracture treatment. In addition, he has received royalties from B. Brawn Ltd related to the design and development of an implant used for the internal fixation of intracapsular hip fractures. R.G. is Honorary Secretary Elect of the Association of Anaesthetists of Great Britain & Ireland (AAGBI). He is a member of the Editorial Board of Anaesthesia. He was the chair of the AAGBI Hip Fracture Guideline Group and founded the Hip Fracture Perioperative Network. S.M.W. is a Council member of the Age Anaesthesia Association, research co-ordinator for the NHS Hip Fracture Anaesthesia Network, represents the Hip Fracture Perioperative Network to the National Hip Fracture Database, sits on the AAGBI Hip Fracture Anaesthesia working party, and provided expert advice to NICE concerning ‘The management of hip fracture in adults’ (CG124). He is an editor of Anaesthesia. C.G.M. is the chair of the British Orthopaedic Association Trauma Group and a member of the Department of Health Fragility Fracture Board. He has received honoraria from Smith and Nephew in the past 2 yr. He is on the International Editorial Board of Injury.

Funding

No external funding was used for this study.

References

1
White
SM
Griffiths
R
Projected incidence of proximal femoral fracture in England: a report from the NHS Hip Fracture Anaesthesia Network (HIPFAN)
Injury
 , 
2011
, vol. 
42
 (pg. 
1230
-
3
)
2
Lawrence
TM
White
CT
Wenn
R
Moran
CG
The current hospital costs of treating hip fractures
Injury
 , 
2005
, vol. 
36
 (pg. 
88
-
92
)
3
Maxwell
MJ
Moran
CG
Moppett
IK
Development and validation of a preoperative scoring system to predict 30 day mortality in patients undergoing hip fracture surgery
Br J Anaesth
 , 
2008
, vol. 
101
 (pg. 
511
-
7
)
4
Wiles
MD
Moran
CG
Sahota
O
Moppett
IK
Nottingham Hip Fracture Score as a predictor of one year mortality in patients undergoing surgical repair of fracture neck of femur
Br J Anaesth
 , 
2011
, vol. 
106
 (pg. 
501
-
4
)
5
Moppett
IK
Wiles
MD
Moran
CG
Sahota
O
The Nottingham Hip Fracture Score as a predictor of early discharge following fractured neck of femur
Age Ageing
 , 
2012
, vol. 
41
 (pg. 
322
-
6
)
6
Wiles
MD
Whitely
JD
Moran
CG
Moppett
IK
The use of LiDCO based fluid management in patients undergoing hip fracture surgery under spinal anaesthesia: neck of femur optimisation therapy-targeted stroke volume (NOTTS): study protocol for a randomized controlled trial
Trials
 , 
2011
, vol. 
12
 pg. 
213
 
7
Association of Anaesthetists of Great Britain and Ireland
Management of proximal femoral fractures 2011
Anaesthesia
 , 
2012
, vol. 
67
 (pg. 
85
-
98
)
8
The National Hip Fracture Database
The National Hip Fracture Database National Report 2010
 , 
2010
London
The National Hip Fracture Database
 
Extended version
9
White
SM
Rashid
N
Chakladar
A
An analysis of renal dysfunction in 1511 patients with fractured neck of femur: the implications for perioperative analgesia
Anaesthesia
 , 
2009
, vol. 
64
 (pg. 
1061
-
5
)
10
Department of Health Payment by Results Team
Payment by Results Guidance for 2010–11
 , 
2010
London
Department of Health
11
Prytherch
DR
Whiteley
MS
Higgins
B
Weaver
PC
Prout
WG
Powell
SJ
POSSUM and Portsmouth POSSUM for predicting mortality
Br J Surg
 , 
1998
, vol. 
85
 (pg. 
1217
-
20
)
13
Musallam
KM
Tamim
HM
Richards
T
, et al.  . 
Preoperative anaemia and postoperative outcomes in non-cardiac surgery: a retrospective cohort study
Lancet
 , 
2011
, vol. 
378
 (pg. 
1396
-
407
)
14
Foss
NB
Kristensen
MT
Kehlet
H
Anaemia impedes functional mobility after hip fracture surgery
Age Ageing
 , 
2008
, vol. 
37
 (pg. 
173
-
8
)
15
Wu
W-C
Schifftner
TL
Henderson
WG
, et al.  . 
Preoperative hematocrit levels and postoperative outcomes in older patients undergoing noncardiac surgery
J Am Med Assoc
 , 
2007
, vol. 
297
 (pg. 
2481
-
8
)
16
Carson
JL
Terrin
ML
Noveck
H
, et al.  . 
Liberal or restrictive transfusion in high-risk patients after hip surgery
N Engl J Med
 , 
2011
, vol. 
365
 (pg. 
2453
-
62
)
17
Serrano-Trenas
JA
Ugale
PF
Cabello
LM
Chofles
LC
Lazaro
PS
Benitez
PC
Role of perioperative intravenous iron therapy in elderly hip fracture patients: a single-center randomized controlled trial
Transfusion
 , 
2011
, vol. 
51
 (pg. 
97
-
104
)
18
White
SM
Griffiths
R
Moppett
IK
Type of anaesthesia for hip fracture surgery—the problems of trial design
Anaesthesia
 , 
2012
, vol. 
67
 (pg. 
574
-
8
)
19
Vandenbroucke
JP
Why do the results of randomised and observational studies differ?
Br Med J
 , 
2011
, vol. 
343
 pg. 
d7020
 

Comments

0 Comments