We wish to thank all the correspondents for their interest in our paper 1, and for their valuable comments. We agree with O'Brien et al. that reasons for persistently similar annual 30-day mortality rates after hip fracture surgery need to be looked at more closely. Our paper, together with data from the recently published Anaesthesia Sprint Audit of Practice (ASAP) 2, confirm earlier work 3 suggesting ongoing wide variations in practice and resistance to evidence-based change. Of course ‘evidence-light’ guidelines are imperfect, but they do provide an informed standard of care to which to aspire. Until we accumulate further evidence about the population with hip fracture, we have a lot to learn about care quality improvement from high volume/low mortality units such as the Royal Victoria Infirmary, that aggregate marginal gains and practice continuous improvement methodologies. Perhaps calculation of a theoretical ‘lowest mortality’, based on a similar demographic without hip fracture, might help focus practice away from ‘how we've always done it’ to ‘what should we aim for?’. We accept that clinicians may disagree with the aims and standards used in ASAP, but propose that any major quality improvement programme has to start somewhere: we have argued for too long about whether general and/or spinal anaesthesia is best for patients, rather than focusing on the future and identifying safer and better-tolerated anaesthetic techniques. The recommendations made in ASAP are based on sound measurement against identified standards, and ASAP represents the largest prospective audit of practice in hip fracture in the UK to date. Our paper did not attempt to address the issue of optimal timing for hip fracture surgery, which has significant humanitarian, medical and financial implications. The results of a planned Canadian ‘Hip Attack’ study should provide further evidence about any benefits of expedited surgery 4. In response to Xue et al.'s letter: the National Hip Fracture Database (NHFD) collects some of the variables stated as potentially ‘confounding’, and we would direct readers to the 2013 National Report 5 and NHFD website (www.nhfd.co.uk), where details of these can be found. We have previously discussed the difficulties inherent in conducting randomised trials involving patients with hip fracture, and the relative equivalence of data yielded from large observational studies 6, but Xue et al.'s point about data adjustment for confounding variables is an interesting one. These are normally adjusted for by collecting large volumes of data, performing univariate regression to calculate the odds ratio of a defined outcome, and then performing multivariable logistic regression to combine the effects of multiple odds ratios in order to identify any potentially attributable remaining difference between interventions. However, this method, frequently used in large database studies, is not without problems. For example, there may be categorisation errors in collecting data: ‘age’ is a reasonably definite variable, but ‘congestive heart failure’ or ‘cardiac history’ may be more problematic, requiring elaborate definitions (if available) that overcomplicate accurate data field completion. Recording error further distorts the dataset. Furthermore, the ‘confounding variables’ identified in analyses that adjust for them are often not justified as potential variables by reference to primary research, may themselves be associations rather than causations, and may be ‘double-counted’ (e.g. cardiovascular disease, one confounder, which increases with age – another confounder). Most importantly, extrapolation of group analyses nationally requires that the group, and its prevalences of constituent confounding variables, is similar to the national population. Anecdotally, cross-reference checks performed on the co-morbidity data presented in ASAP revealed virtually identical data to those recorded in Brighton, suggesting that the NHFD, using the same method of collection as ASAP, could be used to collect and analyse data for confounding variables in future. Indeed, this is one of the aims of the intended secondary analysis of ASAP data, namely to validate the Nottingham Hip Fracture Score as a tool for more accurately controlling for confounding variables in future observational hip fracture studies in the UK; interestingly, age and ASA status (a proxy for confounding co-morbidity), analysed as confounding variables in our study, account for half the weight of this score. Furthermore, the mode of anaesthesia was relatively evenly spread nationally, suggesting that the prevalence of confounding factors in such a large sample is also likely to have been relatively evenly spread, reducing the effect of these on outcome. Ironically, spinal anaesthesia appears to be slightly more frequently preferred for older, ‘sicker’ patients, suggesting that anaesthetists consider it safer for these, and prompting the question of why it is not routinely preferred for younger, fitter patients. Anecdotally, older, ‘sicker’ patients are less likely to receive sedation for surgery under spinal anaesthesia, again presumably for reasons of safety, and (with reference to Fleming's letter) the effect of sedation on outcome is an aim of secondary ASAP data analysis. Although Chesser disputes our findings by re-categorising combined general/spinal anaesthesia within the general anaesthesia group to show a significant reduction in 30-day mortality after general anaesthesia, we reaffirm our view that differences in outcome between modes of anaesthesia cannot be meaningfully analysed further in future unless the confounding effects of neuro-active drugs (i.e. endpoint-adjusted doses of general anaesthetic drugs and sedatives) are accounted for. There are clearly units that predominantly use general anaesthesia with excellent results; equally there are units using predominantly spinal anaesthesia that also get excellent results. Whatever the difference might be between current modes of anaesthesia, it seems to be small compared with the differences between units. We agree with Chesser's second point, that cemented arthroplasty is associated with better long-term outcomes and reduced 30-day postoperative mortality, and we support its use. It was our intention merely to alert anaesthetists to the increase in 24-hour mortality possibly related to bone cement implantation syndrome, which ASAP appears to confirm is of significant intra-operative prevalence. Given the evidence for improved long-term outcomes with cemented hemiarthroplasty we are encouraged by reports of units that have implemented formal protocols in an attempt to reduce the impact of intra-operative instability around the time of implantation. Finally, we can only speculate on Ford's question about why spinal anaesthesia appears to be preferable for similar hip surgery procedures performed electively. Perhaps the very low 30-day mortality recorded after elective surgery is affected by anaesthetic technique in the younger, fitter elective population, whereas this effect signal is lost among the more prevalent confounding variable ‘noise’ associated with the 20-fold higher mortality after hip fracture in the more elderly, emergency population. Perhaps data are more accurately recorded for elective compared with emergency hip fracture surgery. Perhaps spinal anaesthesia has a beneficial effect in some procedures, e.g. reducing blood loss and deep vein thrombosis in total hip arthroplasty, compared with others, e.g. hemiarthroplasty, where the risks of these complications is lower. Clearly, much work remains to be done. By and large, we agree with most of the comments made. The importance of accurate data collection cannot be overstated. Our results should be interpreted in the context of other research, and we need to learn from those units that achieve excellence. However, we should not ignore data just because they don't agree with our currently held views.