Now the TiLLI project has been completed, I have been asked to give a late breaking abstract presentation at the upcoming annual scientific conference of the Royal College of Emergency Medicine (RCEM), to discuss the results.
I suppose the first thing to say, is that this is a good news story for RCEM supported research and ties in nicely with the college research strategy 2020. I worked on this topic initially as a trainee with my colleague Cath Roberts – we were supported to develop some national best practice guidelines for the college on this difficult area. That work highlighted the gaps in research. I took those gaps to the next college clinical studies group meeting, to highlight the research need and seek mentorship and support. As a direct result, the topic found its way into the top 15 James Lind Alliance research priorities and I then worked with the team at the University of Sheffield to secure funding, which included ring fenced academic time for me. The results achieved today, and my career progression, would not have been possible without that help, mentorship and support so I think this is a nice time to thank the research committee of RCEM and the University of Sheffield. With all that support, you’d think we could have come up with a better acronym…
But of course, this project is not about acronyms. It’s about David. David’s story is accessible online and I don’t want to repeat it all here. But in brief, David injured his ankle. It was immobilised in plaster, and he was sent home. He wasn’t assessed for his risk of Venous Thromboembolism (VTE), nor told about the potential complications of immobilisation. He developed a variety of symptoms, and was eventually diagnosed with a saddle pulmonary embolus requiring thrombolysis on intensive care. Despite a lot of post event complications, David made it home alive. Many people is a similar situation, unfortunately do not. Perhaps these are isolated cases? Maybe. But you don’t have to look very far to see multiple other examples. And there is clearly increasing awareness of the issue in patient groups.
But these are anecdotes. And we are scientists. So what is the actual risk. Well, we talked about this 2 years ago at the RCEM conference in 2017 if you remember. if you look closely at recent RCT data you can see a fairly consistent level of symptomatic VTE emerging in the control groups, of around 2%. 1 in 50 people. And these are trial participants, subject to exclusion criteria, often informed and health conscious. This is therefore likely to be an underestimation. Now let’s say we have a therapy that could halve that risk – from 2% to 1%. An ARR of 1%. A NNT of 100. Not that impressive – but what would that mean, given our population affected? We send home approximately 70,000 patients in plaster immobilisation every year in this country. As such, we could be preventing 700 VTE events with this therapy. We get around 350 VTE events per year at my trust. Therefore, 700 therefore seems like a lot to me.
But does it seem like a lot to you? Would that treatment be cost effective? Are there risk factors that could perhaps identify those at higher risk, so we can tailor therapies and maximise benefit? Are there risk assessment models that are reliable, valid and reproducible that can optimise use of this therapy? We thought those were the 4 questions most in need of answering on this topic. And we didn’t think this needed a new randomised controlled trial, to add to the other 10 or so. We thought this needed careful secondary research, to synthesise information, assess clinical outcomes and cost effectiveness based on that information and construct a decision analysis model to highlight how those outcomes and costs might vary, depending on different risk assessment strategies. So off we went….
Question 1 – Clinical Effectiveness
So, Question 1. Which types of VTE prophylaxis have been studied? And do they work? Many of you will have seen the Cochrane review updated in 2017 to reflect addition of the POT CAST study to the trial literature. Did we need to repeat it? Well, it’s always worth going back to the primary data when something is contentious. There are also increasing options for thromboprophylaxis in these patients – aspirin has made a comeback in recent NICE guidelines and the DOACs are increasingly all the rage. We wanted to be sure we had looked for everything.
And we also wanted to perform a network meta-analysis of these options – this is slightly different to a standard SR/MA, as it pools therapies and allows indirect comparison. We chose to lump together all LMWH agents, and compare these to fondaparinux (which is a synthetic factor Xa inhibitor) and to aspirin or placebo. Now, lumping aspirin together with placebo is contentious here. But in our defence, we found limited data and so would not be able to create a single node for this. We also made note of NICE guidance at the time, which explicitly stated not to consider aspirin as prophylaxis against VTE. As we got funded and started, NICE guidance got revised. Just when you think you have everything sorted…..
So what did we find? Well, we can’t quite reveal this as it is under review at present. But I think it would be reasonable to say we found similar data to the Cochrane review from 2017. The odds of symptomatic VTE in that review were significantly reduced with thromboprophylaxis when compared to placebo, with an Odds ratio of 0.40 (95% CI 0.21 to 0.76). Episodes of major bleeding were in single figures, within a population of >6000 cases.
What does this mean? We’re not in a betting shop after all. I am sure you are thinking, what is the ARR, the NNT etc..! Give me something I can work with and explain to my patients. Well, as I sat down with the statistician involved on this to go through these results, I learned a lot. It’s actually fairly inappropriate to try and collate raw percentages from trials like these. Baseline incidence rates are likely to vary between populations, and so compiling the event rates from trials can lead to significant skew. You will notice that Cochrane and others always present forest plots and odds ratios when reporting meta-analysis data. This is why, and there are multiple articles in the literature that support the approach to present relative, rather than absolute measures of treatment effect. These relative measures should theoretically be consistent across populations. If the baseline risk is 2% and the relative risk is 0.5, it stands to reason that the absolute risk in the intervention group should be around 1%. But if the baseline risk is 10% (in a high risk subgroup let’s say), then all things being equal, the risk in the intervention group for this population would be more like 5%. Clearly, adding these ARR values together in one meta-analysis is not going to be without issue.
Question 2 – Individual Risk Characteristics
Next, are there any individual characteristics within this broad population that signify increased risk. We love decision rules/risk assessment methods in EM, but of course these are predicated on the idea that different risk factors convey differing levels of risk. What if they don’t? What if one thing outweighs them all? What if we only had one question to ask at risk assessment? Sounds good to me…
In an ideal world, that’s how all studies looking to construct a decision rule should begin (the TRIPOD checklist is an excellent place to start, if you are thinking about doing this). A systematic review to look at the characteristics that have univariate and multivariate association with the outcome. The multivariate factors are included in a rule, weighted by influence and then formal validation attempted. We often skip straight to the end here, which is naughty.
We attempted to answer this question through a further systematic review. But this one was a mess. We found a lot of evidence, but the vast majority of it was low quality precluding any form of real synthesis. All trials were at moderate to severe risk of bias. Multiple designs were assessed, including logistic regression analyses, non parametric test and descriptive analyses. A real apples and oranges affair.
All we could really do therefore, was highlight consistency across studies. When we looked at this, we found advancing age to be a fairly consistent characteristic for increased risk across 10 studies, with an odds ratio ranging from 1.05 to 3.08. Injury type was highlighted across 5 studies, with OR ranging between 1.88 to 12.7. BMI over 30 was on the fence, highlighted as an increased risk in 4 studies, but with no reported risk in 3 studies. OR ranging from 1.2 to 17.2. This work has been published open access already.
Of interest, following our work on this project the POTCAST authors have also published a post hoc analysis of their RCT cohort, looking to evaluate which risk factors were most associated with subsequent development of VTE. They report a relative risk of 3.8 (95% CI, 1.5–9.4) and 2.4 (95% CI, 1.0–5.6), for BMI>30 and previous history of VTE, respectively. They also report an elevated odds ratio of 2.7 [95% CI, 1.0–6.9]) for patients with an injury requiring surgical intervention, which fits alongside our findings of injury type affecting risk profile. The authors found no particular group stood to benefit from prophylaxis in this analysis, but this result remains subject to the described limitations of the original trial, and the low event rates within.
Question 3 – Risk Assessment Models
So, whilst we shouldn’t really be skipping ahead to risk assessment models, as we haven’t really defined a consistent set of key variables that affect risk, the literature has, and so we must. We looked via systematic review for any mention of risk assessment tools in this population and tried to bring together measures of internal and external validation, to determine the usefulness. Again, most of the studies we encountered were at high risk of bias. But, we did manage to compile a list of 7 RAMs that purport to help in this situation.
There were 2 main points to note from these results. Firstly, there are clearly lots of scores out there. Second, none have been validated properly. A single paper used case control methodology to develop a dataset of 100 patients with a 50% prevalence of VTE and then retrospectively applied these scores. GEMNET appeared to show a sensitivity of 86.5% and a specificity of 4.76%. Plymouth had a much better specificity, but at the cost of sensitivity – Sens 57.1, Spec of 52.4. Within the index paper, Receiving-operator characteristic (ROC) analysis for the L-TRiP score showed sensitivity of 92.6% and specificity of 39.7% using a threshold score of eight. This is much more like it, although further attempts at external validation within the POTCAST cohort suggested this does not perform as hoped.
Since TiLLI has been submitted, we have a new contender on the block as well. This is the Trauma, Immobilisation and Patients characteristics score, published by a similar group of authors to those who derived the LTRIP score. This one was derived by Delphi expert consensus and consists of 30 variables. Thirty. We also performed an expert consensus exercise as part of TiLLI, using orthopaedic surgeons, haematologists, emergency physicians and thrombosis clinicians. We reached consensus on 13 risk-predictors in lower limb immobilisation due to injury. Just goes to show…
In a retrospective database validation, this score showed a sensitivity of around 90% and a specificity of 30%. A subsequent prospective observational cohort study over 2017 was recorded in the same paper, which went on to try and validate this score further. However, this study only recorded proportional risk stratification. Over 35% of these patients got some form of prophylaxis, so commenting on the outcomes (when they have been treated anyway) is clearly significantly flawed.
Question 4 – Show Me The Money
And then the money. The big bucks. Is it cost effective? And if so, what is the most cost effective way of addressing the problem. To look at this we used a decision tree followed by a markov model which involves estimating costs for every possible outcome, then using evidence to estimate proportions of patients who will have that outcome, then totting up the figures. The headline stats? Thromboprophylaxis in lower limb immobilisation has a high chance of being cost effective at the standard NICE threshold of £20k/quality adjusted life year, even if you give it to everyone. If you give it to people based on a decision rule, then you potentially improve that cost effectiveness further. Given there was sensitivity and specificity data available for the LTRIP score, we ran the same model at various thresholds. A threshold score of 8 or 9 seemed to be the sweet spot. This work has been published with follow up articles in review at present.
One of the unexpected interesting outputs from the cost effectiveness analysis, was analysis of drivers behind it. Strangely, this was not really death or serious disease. Mortality outcomes were very low in the literature and fairly even across groups. The key financial drivers were those of index DVT and long term outcomes, such as post thrombotic leg, with associated costs. The important point here for practice, is that even by introducing this therapy you may not save lives. And the coroner needs to be aware of that.
So that’s the project. Yes, thromboprophylaxis appears to be clinically effective at reducing the odds of symptomatic VTE. Only a few individual characteristics seem to discriminate risk, such as age, BMI, injury type and possible FH. There are now at least 8 published RAMS, of which 4 seem reasonable to use, but limited validation studies of any. And the intervention appears to be cost effective, and probably more cost effective if you use a RAM that works.
There are still a few key questions that we get asked about on a regular basis. Can you use DOACs for this indicaton? Does everyone need blood tests prior to starting therapy? Does this evidence apply to splints as well as casts? Which RAM is best? Well, we certainly have our take on all those questions, but we didn’t find much evidence to help. A further review article will hopefully be published shortly in the EMJ which addresses all these points and provides an overview of this work, in addition. I don’t want to ruin the punchline for this. So I won’t.
Where next? Well, there is always more work to be done. I think our priority is to amend the RCEM guidelines to reflect this new evidence, and to produce some dissemination aids through national information leaflets, learning resources. Our patient group had a very interesting take on some of these latter points, and were hugely keen to do awareness raising – they also interestingly wanted the option to be able to purchase private prescriptions for TP, even if the healthcare system they were in deemed to be low risk. I’m not sure how I feel about this yet. But I am clear from working on this project, that information and communication of risk, including ways to mitigate it, is absolutely key. Many of the bad outcomes we see in this situation arise because people do not understand the risk of VTE, and do not present to hospital early when they develop signs or symptoms. We can change that. Today.
Lastly, I would just like to thank all my co-investigators on this project, who worked hard to deliver TiLLI. They are a fantastic group of academics at Sheffield and I remain grateful for the collaboration and support.
Well done for sticking with me if you have got to the bottom. But hey, don’t be too proud of yourself. The full TiLLI report is >80,000 words. It will be out soon. Let me know when you have read that, and I’ll send you a sticker.