Principal findings
The evidence identified in the scoping review mostly comprised small, retrospective, uncontrolled group studies (such as case series) or case report studies. Although the published studies covered a wide range of treatments for extravasation injuries, few studies formally graded injury severity at baseline and the results sections of most studies were brief and lacking important information. Furthermore, there was considerable clinical heterogeneity across study populations in age, types of infusate, injury severity, location of injury and the time gaps between injury identification and subsequent treatment. Differences in results across studies might be a reflection of variation in one or more of these parameters, rather than differences in treatment effect. Consequently, uncertainty exists regarding which treatments may be the most promising, particularly with respect to how to treat earlier stage injuries (i.e. injuries which have not become necrotic). Notwithstanding the study limitations, some of the better evidence, in terms of study size and a prospective design, related to studies of flush-out techniques, which appear to be quite promising treatments. However, the effect of prior infiltration with hyaluronidase before flush-out is unclear.
The use of a scoping review, rather than a full systematic review, to assess the literature was justified on the expectation that any review was very unlikely to produce evidence robust enough to allow treatment recommendations to be made with sufficient confidence. Scoping reviews are broader and more exploratory in nature than full systematic reviews. They are often undertaken when an evidence base is expected to be either very large or have important limitations; the latter being the case for this review.
The NHS survey results showed that, although most units (82%) had a written protocol or guideline for treating extravasation injuries, a staging system for grading severity of extravasation injury was included in just over one-third of protocols or guidelines. Almost all responders indicated that peripheral lines were the access site most associated with extravasation injuries. In neonatal units, parenteral nutrition caused the largest proportion of extravasation injuries, whereas in principal oncology/haematology units, the largest proportion of injuries was due to the extravasation of vesicant chemotherapies.
The survey showed that the most frequently used intervention approaches were elevation of the affected area and analgesics. The results also revealed that, in most units, warm or cold compresses were rarely or never used. In neonatal units, there was notable variation regarding the use of occlusive dressings, ranging from always being used (8% of responses) to never being used (31% of responses). Variation in the use of saline irrigation (or wash-out), either with or without hyaluronidase, was also evident; these interventions seem to be usually or sometimes used in around half of the neonatal units, although they are never used in around one-third of units. Results for principal oncology/haematology units and PICUs were broadly similar to the neonatal unit results.
When asked about a future research study, most responders (65%) thought a RCT would be viable, although these results varied by setting: 83% of neonatal unit responses, 33% of principal oncology/haematology unit responses and 0% of PICUs. Almost all the responders who thought a RCT was viable mentioned one or more of the following treatments when asked which treatments they would most like to see studied: saline irrigation/wash-out, hyaluronidase and conservative management. Of those who thought a RCT was not viable, the reasons included too many variables which could affect outcomes, timeliness of treatment when using randomisation, low numbers of patients and unwillingness to deviate from current practice.
Strengths and limitations
The scoping review was performed using systematic, reproducible, transparent and robust methods. Our bibliographic database searches were comprehensive to allow identification of all relevant published studies and searches were also made to identify any unpublished studies. These methods minimised the possibility of publication or language biases affecting the review. The possibility of reviewer errors and biases affecting this review were minimised by performing review processes in duplicate. The sample for our NHS survey was large and diverse enough to be representative of NHS staff who treat extravasation injuries. The main limitation of the scoping review related to the evidence identified. Most studies were limited in helping to evaluate which treatments might be best and also in presenting ideas regarding which direction future research studies should take.
Patient (parent) and public involvement
Extravasation injuries are quite rare events and patient groups specifically in this area do not exist. However, one of our clinical team discussed this study with the parents of an infant who had suffered a severe extravasation injury. This proved to be useful in informing the survey content. For example, in the section of the survey where we asked questions about how frequently specific interventions were used, there was initially no question about use of analgesics. The importance of pain relief was emphasised by this parental input and, consequently, a question on analgesic use was added to the survey.
Future research on treatments for extravasation injuries
The survey results indicated that, across NHS neonatal units generally, there was optimism about randomised trial feasibility, although this was not the case for other types of units. The scoping review studies yielded little information about future research. Only three of the comparative or non-comparative studies mentioned issues relating to research study design.31,39,40 In two studies, this was very brief: mention was made of the need for a prospective controlled study for confirmation of findings,40 and that a randomised trial was not possible because of the low incidence of injuries.39 The authors of the third study commented, somewhat vaguely, that optimal management is uncertain because of ethics considerations limiting controlled research, although they added that a centralised register of extravasation events would be a useful means to monitor, assess and review outcomes.31 One of the systematic reviews (of chemotherapy extravasations) also discussed the challenges of undertaking further research, noting key issues which may preclude the use of a randomised design.164 These included the sporadic occurrence and low incidence of extravasations, and the complexity involved in controlling the many extraneous variables associated with extravasations: age, sex, comorbidities, type of infusate, site and volume of extravasation and time to intervention. The authors commented that the sample size required to properly control for these factors will probably be prohibitively large. Notwithstanding the discussion in this systematic review,164 the very limited insight on future research provided by the scoping review studies means that further exploration is needed. This is in relation to decisions regarding the study design, treatments to be studied, population to be studied and the outcome measures needed. Such decisions should not be made independently of each other. For example, a randomised trial would probably involve fewer, and more defined, interventions and a narrower population than an observational database study. As these two study designs, randomised trial and a database study, appear to be important options for any future study, the key issues relating to each will be discussed to help inform a decision on future research.
Randomised trials
Exploration of feasibility issues
Treatment delays and selection bias
Extravasation injuries require urgent treatment. The time gap between identifying and treating an injury is a key factor which determines important clinical outcomes. However, the process of recruiting and randomising patients into a clinical trial often results in treatment delays. Initial delays may arise when ascertaining whether or not a patient is eligible to participate. For those who are eligible, further delays may arise from the randomisation process. This is most likely to occur where the procedure of allocating treatment is performed in real time. For example, some delay would be inevitable when contacting a central randomisation service provider, either via the internet or by telephone. The scoping review identified that the time delay between injury identification and treatment varied widely across studies. Important questions when considering trial procedures are ‘what is the typical time gap between identification and treatment of an injury in the NHS?’, and, ‘is this short enough to accommodate a delay due to randomisation?’ (i.e. would the randomisation delay be acceptable or difficult to justify clinically?). Arguably, the most impressive results identified in the scoping review were from a study in which neonates with stage III and IV injuries were treated within just 10–30 minutes of injury.4
A frequently used method of randomisation is the use of sequentially numbered, opaque, sealed envelopes containing randomly generated treatment allocations. Adoption of this method might minimise such delays, but this approach has been demonstrated to be prone to investigator selection bias. Reports of surgeons opening envelopes in order to subvert randomisation, and of trials using sequentially numbered, opaque, sealed envelopes being more likely to show statistically significant treatment effects than trials using more secure allocation methods, suggest that envelope-based methods should be used cautiously.168 The risk of bias might only be reduced when the personnel with access to the envelopes are different from those actually recruiting participants. Of course, having an added layer of trial personnel involved in the recruitment process might itself add to treatment delays (and also to the administrative cost and burden of the trial). However, these concerns may have more limited relevance to trials performed in emergency or urgent care settings, where there is some evidence to suggest that the risks of selection bias may be low.169
In trials where interventions exist in discrete packs and look indistinguishable from each another, or can be made to look so, a different type of time-saving method may overcome concerns about using envelopes. In this approach, randomisation identification labels which are pre-coded would be attached to ‘extravasation kits’ which would then be placed in order and allocated to patients sequentially. The pre-coded labels would be meaningless to the investigator, who would have only very restricted access to the randomisation coding system (e.g. where a serious adverse event was suspected). Heat-sealed bags could be used to reduce the risk of tampering and subverting the randomisation sequence. This type of method should obviate concerns about both selection bias and delays in receiving treatment as a result of randomisation because the randomisation process would simply involve the next kit being taken, used and recorded.
Nevertheless, many interventions cannot be code labelled as identical-looking, discrete packages, for example, different types of debridement or surgery. In these situations, a real-time randomisation process may be needed, which could lead to treatment delays. A possible alternative here might be the use of quasi-randomisation methods. This typically involves the use of a pre-defined participant or setting characteristic, such as date of birth or day of the month, to determine treatment allocation (e.g. odd days indicate treatment A and even days treatment B). The scoping review identified an old quasi-randomised trial reported by Brown et al.55 which allocated treatment according to calendar month. Quasi-randomised studies often allocate treatments based on day of the month, but, given the scarcity of extravasation events, this idea of using alternate months seems a pragmatic approach to recruitment and treatment allocation. Quasi-randomisation would solve the time delay issue, but the selection bias issue would remain; however, as already noted, the risks of selection bias appear to be low in urgent care trials.
The requirement for urgent treatment also raises the issue of consent to participate, as this might also lead to treatment delays. However, in urgent care settings, a case can be made for using deferred consent, rather than prior consent. Deferred consent enables children to be included in trials without prior informed parental/carer consent, but requires such consent to be acquired as soon as possible for continued trial participation.
Recruitment
Perhaps the most important barriers to successfully executing a randomised trial are those which hinder the accrual of enough participants to yield meaningful and reliable results. Extravasation injuries are quite rare events which are also subject to variation, particularly in terms of patients (ages, comorbidities), causes (infusates), injury sites and severities and the speed at which injuries are detected and treated. Consequently, careful consideration would be needed when devising trial eligibility criteria to enable the recruitment of both a sufficiently homogeneous sample of participants and a sample which would be large enough to minimise the impact of chance differences across treatment groups in any of these factors. Failure to do so would increase the risk of false-positive trial results; small trials are more prone to yielding chance results than larger trials.
It is likely that a large number of participating hospital sites would be needed to allow adequate recruitment. Although the scoping review focused on children, it also involved an informal assessment of a broader range of papers on extravasation treatments in adults; however, only one randomised trial was identified.170 The trial, reported as a conference abstract, compared different cooling treatments for doxorubicin extravasation; after 7 years, only 37 patients had been randomised (the study began in 1987 and ended in 1994).171 This example serves to highlight the recruitment difficulties which might be encountered in any future RCT in children.
The discontinuation of randomised trials wastes research resources and also raises ethical concerns. A study of 1017 RCTs172 found that 25% were discontinued, the most frequent reason being poor recruitment (occurring in 10% of the 1017 trials). Trials discontinued because poor recruitment achieved a median percentage of target sample size of 41%. Trials with investigator sponsorship (compared with industry sponsorship) and trials with smaller planned sample sizes were at higher risk of discontinuation because of poor recruitment. A UK HTA study of 114 multicentre RCTs,173 which were funded by the Medical Research Council or the NHS (HTA), found that less than one-third of trials recruited their original target within the time originally specified and around one-third had extensions. The following factors were observed more frequently in trials that recruited successfully: having a dedicated trial manager, being a cancer or drug trial and use of treatments only available within the trial setting. Results from a survey of 181 principal investigators in a large US paediatric hospital174 found that the method of recruitment appeared to be the only significant and independent factor associated with achieving 80% or more of target enrolment; protocols that used recruitment in person were 4.6 times (95% confidence interval 1.3 to 15.9; p = 0.02) more likely to achieve 80% or more of their target enrolment when compared with those that used other recruitment methods. Utilisation of electronic data recorded in clinical practice databases or registries could reduce recruitment problems; therefore, randomised registry trials, sometimes referred to as pragmatic randomised electronic point of care trials, may be a useful approach and will be discussed in Database studies.
Alternative trial designs to maximise recruitment
Small sample sizes are a common problem in paediatric trials, which, consequently, are often insufficiently powered to detect true treatment effects. Innovative approaches may, therefore, be considered to overcome this issue.175 Sample size estimation can be a particular challenge when designing paediatric trials; this would be an issue for any trial of extravasation treatments because the effectiveness data available to inform such calculations will be minimal. However, in trials using a ‘sequential’ design, the sample size at the end of the trial is not known at the beginning; trial stopping rules are defined based on the accumulated data and, therefore, trials can end on the basis of efficacy or futility. This design may be suitable in trials where outcome results are available quickly in relation to the patient recruitment rate. Analyses from a systematic review of paediatric sequential trials176 (24 were published between 1963 and 2005) indicated a median reduction in sample size of 52 subjects (range −22 to 229 subjects) or 35% (range −42% to 90%) for sequential trials when comparing with a classical fixed sample size approach. Only nine trials reported sufficient information about assumptions to allow calculation of a corresponding fixed sample size. Thirteen of the 24 trials were performed in a NICU setting. Although for any given sequential trial it remains a possibility that the eventual sample size may turn out to be larger than the fixed sample size, it is evident that fewer patients are generally necessary to reach a conclusion, compared with a fixed sample size design, thus providing some ethical advantages. The unknown sample size at the start of sequential trials may be problematic for funders, although some trials pre specify a maximum number of participants.176
Another strategy to address likely sample size problems is responsive–adaptive randomisation: a ‘play the winner’ approach that maximises allocation to the most effective treatment. Outcomes for previous participants affect subsequent treatment allocation probabilities. This design is, therefore, also limited to studies which assess outcomes quite quickly. However, the Food and Drug Administration177 has expressed concerns regarding the magnitude of the risk of bias and the size of the potential bias, and how to eliminate these effects, as they are not yet well understood for adaptive trial designs.
Blinding
A lack of blinding (also known as treatment masking) could be a source of bias in a trial of treatments for extravasation injuries. Although blinding would certainly be possible for some treatments, for example, hyaluronidase injections, where placebos could be used, it would not be viable for others, for example, when comparing saline flush-out with a conservative management intervention. In an extravasation trial of infants, there would be two mechanisms by which a trial might be biased by lack of blinding. First, via systematic differences between the care provided (e.g. cointerventions) to the different treatment groups (i.e. performance biases), and second, where outcomes assessors are aware of intervention assignments (i.e. detection biases). For example, detection bias may occur where it would be possible to distinguish between an injury treated with saline flush-out, which may still have puncture marks, and one treated with conservative management, which will not. With this in mind, it is not helpful that most outcomes in an extravasation trial would be subjective in nature (e.g. scarring, wound healing time). However, the risk of detection bias can be considered to be low if the outcome assessor used is an independent researcher, rather than a trial investigator; the latter may have treatment preferences, the former probably would not. Blinded assessment of photographs may also be a useful way of evaluating wounds over time. The use of parental assessment of outcomes could also be considered. An evaluation of performance bias can be made by recording, and later assessing, any deviations in the care provided which are beyond what would be expected in usual practice.
Treatments, populations and outcomes
The low rates and sporadic incidence of extravasation injuries are important issues to consider in any future research; this was evident from the survey where low incidence of injuries was noted several times as a barrier to a RCT, particularly any trial involving principal oncology/haematology units. As outlined in Chapter 1
Prevalence of extravasation injuries and risk factors, the incidence of extravasation injuries seems to be higher in preterm neonates (particularly those receiving an i.v. intervention at a peripheral site) than in other infants, making this, perhaps, the most viable population for a randomised trial. Restriction to parenteral nutrition infusions should further reduce population heterogeneity, although one survey respondent noted that, even here, there may be significant heterogeneity to overcome. Our survey results indicate that parenteral nutrition infusions at peripheral sites are the most frequent cause of extravasations in neonates. It is unclear, however, whether the preterm neonate population would be large enough for a randomised trial. The severities of injury to include in a trial would depend on the treatments being studied and on which paediatric injury grading scale is adopted.3,15
For a preterm neonate population, the likely treatments to compare could be a specific form of conservative management compared with a saline flush-out technique. This is based on both the scoping review and survey results. The parameters used for saline flush-out can vary, such as the number of puncture points and the volume of saline; fewer puncture points and lower volumes (than is used for older infants) would seem appropriate in preterm neonates.
Although it is medically plausible to expect that prior hyaluronidase injection plus saline flush-out may be more effective than saline flush-out alone, there is little robust clinical research evidence to support this. The magnitude of any differences in outcome between these treatments could be small; therefore, a sizeable trial would be needed to allow such a difference to be demonstrated. Also, the appropriateness of using hyaluronidase in pre-term neonates is unclear. One study included in the scoping review noted that infiltration with hyaluronidase is an invasive procedure and the British National Formulary46 has advised caution in the use of hyaluronidase in infants and to control the speed and total volume of injection.
The degree of scarring following extravasation injury treatment will be an important outcome in any future study. In the scoping review, very few studies quantified scarring outcomes. Several measures exist for assessing scarring, but there is little consensus as to which is the optimal scale or tool to use.178 A gold standard scar scale does not currently exist, although, ideally, such a scale should address cosmetic, functional and psychological sequelae.179 Outcomes which involve parental evaluation, in addition to clinician evaluation, should also be considered for certain assessments, including scarring. The reliability and validity of outcome measures would need to be established for use in neonatal populations.
Database studies
Exploration of feasibility issues
A prospective database or registry study may be appropriate where a randomised trial is not considered to be viable for practical reasons. A database study would likely result in a larger sample than would be obtained in a RCT because of simpler recruitment processes and a broader population from which to recruit. The downside, or trade-off, with adopting such a non-randomised approach to comparing interventions is that the results would inherently be less reliable than those of a RCT; although, as previously discussed, RCTs which do not recruit enough participants may also produce unreliable results. The most important methodological difference between randomised and non-randomised studies arises as a result of confounding, caused by selection bias, which will often be encountered in non-randomised studies. Methods exist to adjust for confounding, such as regression analysis, propensity scoring, instrumental variables, stratification and matching, but it is unclear which methods are most appropriate in any given circumstance.180 When aiming to generate believable estimates from non-randomised studies, it is necessary to identify the important confounding factors which need to be measured validly and precisely.181 The risk of confounding arising from some factors might also be reduced by narrowing eligibility criteria. This would, however, limit the generalisability of the study results to wider patient groups.
However, we have already noted the extent of variation in important baseline characteristics in the studies identified in the scoping review: consideration would be needed regarding adjustments for variation in comorbidities, injury sites, methods of delivery, duration of i.v. therapy, the amount of fluid extravasated (very difficult to estimate) and the speed at which injuries are detected and treated. It could, perhaps, be argued that the latter four factors might be covered by use of a staging system for grading the severity of injuries. However, our survey results indicate that most units do not use a staging system to grade injury severity. Furthermore, variation is likely across units which do use such a system, since different approaches to grading injury severity have been published (see Chapter 1). An observational database study would also inevitably result in variation in the treatments given, even where they might be considered to be the same. For example, for flush-out techniques, variation exists in the number of puncture points made and the volume of saline used.
Nevertheless, a database study would be considerably cheaper to undertake than a randomised trial, especially if it were to utilise existing relevant database facilities such as the UK National Neonatal Research Database (NNRD). This database holds data entered by UK neonatal professionals. However, it does not currently routinely record data on extravasation injuries.
Randomised registry trials
If such data were recorded in a database in the future, the possibility of a randomised registry trial (also known as pragmatic randomised electronic point of care trials) could be explored, as the results of a UK survey of neonatal health professionals suggest this approach would be both feasible and acceptable in neonates.182 Randomised registry trials are pragmatic randomised trials performed in usual clinical care conditions, which utilise routinely collected data. Trial interventions fall within accepted professional standards but, as yet, have uncertain comparative effectiveness. Ideally, the recruitment and follow-up procedures would be naturalistic and mimic actual clinical decisions and practices, except for the randomisation process.183
These trials use the registry as a platform for recruitment and trial administration. Theoretically, this approach is appealing because it keeps the best aspects of both RCTs (i.e. robust, unbiased estimates of effectiveness) and registry studies (i.e. larger sample sizes) and, consequently, dispenses with some of the worst aspects (e.g. small samples with limited generalisability and estimates which are inherently unreliable). Importantly, registry RCTs should also be cheaper and quicker to undertake than conventional RCTs. A more detailed discussion of this trial design, including example trials, was published in 2016.184
Some of the challenges to consider when planning a registry RCT include ensuring that the data are of high enough quality, consideration of blinding and the standardisation and adjudication needed for certain outcomes.185 For any extravasation trial some key issues could include achieving a consistent use of a single staging system to grade injury severity when recruiting patients, the degree of standardisation/consistency needed when administering the studied treatments and the choice of outcome measures, which would need to be clinically practicable yet also demonstrate adequate reliability and validity.
Treatments, populations and outcomes
The treatments, populations and outcomes most viable for a randomised registry trial would be the same as those discussed in the previous section on randomised trials. But consideration should also be made regarding which treatments and populations might be studied in a prospective observational (i.e. non-randomised) database or registry study.
Usually, a key advantage of a prospective database study over a randomised trial is that a larger number of treatment approaches and a broader population could be studied. However, in this area of research, such benefits might, in reality, be quite small. Consider, for example, a population such as children receiving i.v. chemotherapy. The scarcity of chemotherapy extravasations, evident from the survey, coupled with the wide variation in infusates, subpopulations and injury treatment approaches, evident from reviews and guidelines, mean that accruing a cohort sufficiently large enough to produce meaningful comparative results may be very difficult. It is somewhat unclear to what extent this might be said for extravasations of other infusates such as calcium, contrast agents and blood. In mitigation, database studies by their very nature are able to accrue data over very many years. A database study might be useful for evaluations where there is less variation in the treatment options available. For example, for injuries which have become necrotic, comparison might be made of outcomes following different methods of debridement: mechanical, enzymatic and surgical.
Although beyond the scope of this review, research is also needed on interventions to prevent extravasation injuries. Ideally this would begin with a systematic review.
Summary
Careful thought is needed when considering any future comparative study of extravasation injury treatments. Decisions regarding the study design, treatments, population and outcome measures should not be made independently of each other. Some of the practicalities involved in undertaking a conventional RCT, such as recruiting adequate numbers, avoiding treatment delays and selection bias, could be difficult to overcome. Although a prospective observational database study would maximise the number of patients recruited, and eliminate concerns about treatment delays, its results would inherently be subject to uncertainty as a result of the likelihood of selection bias.
Several alternatives exist to a conventional RCT design, which still include a randomisation element. Perhaps the most promising is the randomised registry trial, which incorporates many of the best aspects of both conventional RCTs and observational database studies. Although this design is relatively new, and few trials have been performed, its relevance to a trial of treatments for extravasation injuries is worthy of exploration. However, a key relevant database, the UK NNRD, does not currently record data on extravasation injuries. Further issues to be considered in any randomised registry trial of neonates include the lack of a protocol or guideline for treating extravasation injuries in 25% of units, and the absence of the use of a staging system for grading injury severity in over half of the units which do have access to a protocol or guideline.
The low rates and sporadic incidence of extravasation injuries and population heterogeneity are key issues when considering the population to be studied. As such, preterm neonates receiving i.v. parenteral nutrition at a peripheral site may perhaps be the most viable population for any randomised trial. The main treatment candidates receive are a standardised conservative management intervention, saline flush-out without hyaluronidase and saline flush-out with hyaluronidase; the choice of treatment would depend on the injury severity grades chosen to be eligible for inclusion. A number of different methods exist for grading injury severity, with variation likely across the NHS; for example, some units do not formally grade injury severity. A paucity of standardised outcome measures used in previous studies in neonates is also a concern. Outcome measures used in a future study would need to be clinically practicable yet also demonstrate adequate reliability and validity.
