Thinking fast and slow in the evaluation of injury plausibility in child protection
Conflict of interest: The author has provided expert testimony for prosecution and defence in criminal proceedings, child protection proceedings and coronial inquiries. None of this work has been personally remunerated and is undertaken within the service and funding model of Children's Health Queensland.
Abstract
In evaluating injury in children, child protection paediatricians are tasked with determining whether the history given by parents or caregivers is valid with respect to explaining injury causation. This paper summarises metacognition and in particular the dual processing theory of ‘fast and slow thinking’ to explain how complex information in contexts of uncertainty is processed to produce decisions and responses, applied to a child protection context. An example is used that resembles abusive head trauma which focuses on understanding the difference between likelihood and plausibility in forensic interpretations of injury causation.
How Do We Think? A Model for Metacognition
A bat and a ball cost $1.10. The bat cost $1 more than the ball. How much did the ball cost?
While most will respond with 10 cents, the correct answer is in fact 5c. Known as a cognitive reflection test, this was devised by Professor Daniel Kahneman, Nobel laureate, who wrote the sentinel work on metacognition ‘Thinking, fast and slow’ describing the dual processing model of the brain. These two modes of thinking, system I (fast) and system II (slow) have relevance to all human decision‐making and in particular, to science and medicine1 (Table 1).
| System I (fast thinking) | System II (slow thinking) |
|---|---|
| Intuitive, automatic, impulsive | Effortful, deliberate and analytical |
| Without conscious control | Hypothetico‐deductive method of medical diagnosis |
| Based on pattern recognition | Inferential reasoning in forensic medicine |
| Uses associative machinery to form impressions | Requires attention, exertion of effort and control |
| Involves ‘rules of thumb’ – mental‐shortcuts (spot diagnoses) to make quick instantaneous decisions | Usually on standby |
| Driven by need for coherence, which suppresses ambiguity – allows minds to ‘fill in the blanks’ | Develops an argument through logic and checks conclusions |
| Coherence is associative and emotionally driven | Thoughtful and calculated, invokes deliberate memory search |
| Just happens as a product of human mind | Effort and attention are finite resources |
| Imposes causal thinking to components – may not be valid | Inferences involves reasons and arguments, order and sequence are essential |
| Associations just happen, does not draw upon reasoning process | Reasoning must lead logically from premise to conclusions |
Kahneman described common cognitive biases which are systematic errors of thinking that affect human decisions and judgements, to classify errors in decision‐making which have since been expanded on. 2, 3 In medicine, cognitive biases are thought to account for up to 70% of medical errors.4-7 Metacognition involves the capacity for self‐reflection on the process of thinking and self‐regulation in monitoring decision‐making.8 In medicine, well‐developed metacognition can improve diagnostic accuracy.9
Errors with System I and System II Thinking
Most cognitive biases occur using system I mode of thinking. The 10 cent answer to the cognitive reflection test illustrates how, with very little information, it is easy to be confident in the wrong answer. When a difficult judgement is required and a related judgement comes easily to mind, the easy judgement is often used instead of more complex thinking modes. If a person's life depended on it, arguably a person would be less likely to give the 10 cent answer because they would have checked their answer first. The gravity of the situation would have forced them to switch into system II mode of thinking.1, 2
A fundamental rule of human efficiency is the SATO principle which refers to Speed/Accuracy Trade‐Offs.10 Dual processing theory suggests the two systems function in sequence; heuristics are used to immediately solve the problem, and analytic reasoning may (or may not) be employed to alter the original impression. We operate within system I thinking most of the time and system II can, when called upon, act as a filter.
Metacognition is the intentional engagement of system II problem‐solving through reflection and deliberate examination of one's own reasoning.11, 12 System II can also lead to mistakes. As Kahneman noted ‘We make very significant mistakes when we think very seriously’.1 Errors in system II thinking can ‘create paralysis by analysis’, can arise when a person reasons correctly from a false premise, reasons illogically, uses flawed mental modelling (such as using linear reasoning in a complex adaptive system) and from having inadequate content knowledge.12
Is the Explanation We Are Told Believable?
This viewpoint summarises different cognitive mechanisms involved in the assessment of information and considers from my perspective how these are appropriate and adaptive, and where they may introduce potential bias or error in the evaluation of injury in a child protection context. The medical task in child protection is to determine if an injury is explained by the history provided which may or may not be truthful. They may relate highly irregular and uncommon circumstances that do not sound immediately plausible but nevertheless have occurred and have resulted in accidental injury. Making an error in this context has profound implications, in both directions.
Consider this case – see Box 1.
A 10‐month‐old child dies from an alleged unwitnessed short distance fall from a couch from a rapidly expanding large unilateral mass‐effect subdural haemorrhage following an initial period of lucidity followed by coma. Unlike most bridging vein injuries which thrombose and clot, this one has not and has continued to bleed rapidly resulting in brain herniation and death despite emergency neurosurgery and extensive transfusion of various blood products. Extensively distributed retinal haemorrhages are found to the ora serrata and in multi‐layers from the hyperacute intracranial collection, without retinoschisis or folds.
Is this a case of fatal child abuse or could it have been a fatal accident without involving another person?
Possible Biases in Child Protection
A marker of system I activation is when emotion is involved, which activates a neural association network then prepares us to interpret what comes next in a particular way. In child protection the presence of a significant injury in a young child, or parental behaviours/responses when being interviewed in this context can produce an emotionally driven priming effect on the forensic expert which may inadvertently introduce bias.
In child protection paediatrics, involvement begins when the history does not appear consistent with the injury or is unusual. Both situations may elicit emotion and this may motivate potential bias towards a position of disbelief. If this occurs, system I thinking may jump to conclusions that validate that disbelief causing a system I error if the paediatrician has immediately discounted the history and has not shifted into a systems II analysis of the specific facts. These are often only available after thorough investigation. See Box 2.
Applying System II Metacognition to Injury Plausibility
considers case specific established facts from investigation
applied to authoritative evidence (research, consensus publications)
- mechanism of injury of this type – in general then in this case
- how much force is involved – in general then in this case
timing – relevant features from medical investigations
and evolution of symptoms and observable functional effects
to determine if the explanation is plausible
System I thinking is therefore useful to rapidly triage information and make quick decisions including recognising the threshold to involve investigators and to make immediate child protection decisions where the overriding safety of the child is paramount. Systems II thinking should then follow which involves a more thoughtful, deliberative and analytical approach to reach an ultimate conclusion.
Adversarial biases arise when experts either consciously or subconsciously align their thinking with the ‘side’ that has sought their opinion. Experts can be motivated by financial rewards, prestige or derived from belief in a ‘cause’.13 As an example, this can arise when experts fail to include child abuse in their differential (e.g. defence‐only experts) due to sole reliance of system I thinking without shifting into systems II analysis. This has been referred to as ‘denialism in child abuse’, whereby pseudo‐scientific hypotheses are posited as valid conditions to achieve a coherent argument in line with their preferred narrative.14-16 These views are typically driven by emotion which can operate to produce bias in either direction (to invoke or discount abuse), by producing the same conclusion in all submitted cases, regardless of the truth or analysis of the individual facts.13
Expanding Metacognitive Models
The routine expert can only use pre‐existing knowledge to quickly solve routine, familiar or uncomplicated problems, by using heuristic ‘illness scripts’ (diagnostic clues that inform diagnosis). The adaptive expertise model has been developed to further expand on metacognition in medical diagnostics.12 The adaptive expert employs a deep conceptual understanding and can engage in reflection to create novel situations for complicated or unfamiliar problems thus adding to their knowledge base, reasoning capacity and ability to solve cases not previously encountered. Adaptive expertise grows as knowledge and problem‐solving skills grow, allowing the expert to make valid diagnoses not only in simple scenarios but in more complex, uncertain or unfamiliar cases by using reasoning and logic bringing innovation in finding solutions. Adaptive expertise relies on the ability to engage in both systems I and II thinking to problem‐solving by balancing efficiency with innovation, applying foundational knowledge and learned experience to formulate novel solutions.9
Incorporating Metacognition into Child Protection Training
Education can develop system I thinking to expand and improve associative networks which improves recognition of possible cases of child abuse and neglect. By educating about system I ‘illness scripts’, expert intuition develops this associative network allowing the clinician to draw upon a vast knowledge base, improving recognition from pre‐existing information to quickly solve routine, familiar and uncomplicated problems. In child protection, teaching novices about injury patterns and prevalence from epidemiological studies helps them recognise injuries suspicious for an inflicted cause that do not conform to injuries typically sustained as a result of accidents at any given age.17-19
As it is not possible to operate in system II all of the time because the requisites of attention and effort are finite resources, system I thinking needs to be trusted to some extent in initially triaging cases which may, or may not, involve child protection risk. This is where errors mostly arise from cognitive biases. Training can also target system II responses. Through education a person can learn to recognise situations in which system II should take over, thereby avoiding mistakes which can arise with the sole application of system I thinking.12
In medical pedagogy, educational interventions which focus on teaching recognition of common cognitive biases and studies using cases of medical error and debiasing strategies (e.g. use of algorithms, checklists, protocols) that promote application of system II, have been shown in the short‐ and long term to improve residents' critical thinking.20, 21 Learning and practicing strategies to avoid biased thinking are termed debiasing or ‘cognitive forcing’ strategies which work by disrupting the automaticity of clinical reasoning, imposing on the clinician the need to re‐evaluate their initial thought processes through reconsideration of evidence. These skills must be practiced regularly or else inductive reasoning and metacognitive skills wane.12 To help learners achieve adaptive expertise, educators must help them recognise the appropriate use and challenges of both heuristics and analytical reasoning, adapt their approach of diagnosis to the specific clinical scenario and become comfortable with regular encounters of uncertainty. This is particularly pertinent to the landscape of child protection where, in order for novices to learn this they must experience this in the workplace and the workplace needs to use these models within clinical practice.
Reasoning in Child Protection
Adopting sound forensic practices of communicating expert opinion in sequential steps creates the opportunity for system II analysis of specific facts before forensic opinion is finalised for the criminal justice system. The information needed to undertake this task may differ from what is available when the child is initially in hospital in the immediate aftermath of their injury. The evaluation/medical workup should be thorough to avoid risk of confirmation bias (looking only for what you expect to find). Following admission and workup for a suspicious injury an initial report is required in a timely way by statutory agencies to contribute to interim child protection decisions. This must allow for the fact that investigation is at that time, still typically unfolding.
By the time a case is heard in the criminal jurisdiction experts being called by defence legal counsel would have had access to the entire case investigation. It is essential that the child protection paediatrician has considered the same information and objectively considers any further information that may have come to light subsequent to their own involvement and avoids being inadvertently anchored by their initial impressions. What was opined originally may need to be re‐evaluated in this later context, where there is a different standard of proof, focus of argument and procedures applied to the evidence.
Using Systems II to Communicate Injury Plausibility
By the time cases are heard within the criminal courts I would argue that it is rarely possible to rule out child abuse. What is considered by the expert in that context is whether the explanation provided is plausible. Put simply, injury plausibility considers whether the explanation/facts/circumstances could have caused the injury. This falls short of the expert saying what actually did, or did not happen, and conforms with the rules of expert evidence which require the jury or judge (in judge‐only trials) to be the final arbiter of the ultimate issue of guilt/innocence. The court considers the experts contribution alongside all the other facts that have been raised within the proceedings.
By the time a matter is being considered in the criminal jurisdiction a forensic opinion should, in my view, avoid context bias from any prejudicial or positive psycho‐social factors and instead be focused on consideration of biomechanics, case specifics and injury plausibility. This may shift thinking from what is most likely (which is an appropriate way to think when first evaluating a child with an injury) to whether the agreed upon set of facts which forms the history at that point is plausible. This will allow for the correct identification of situations which may in general be unlikely but nevertheless have happened in specific rare situations.
This can be illustrated with the following example: If you went swimming in the ocean it would be extremely unlikely that you would die from being struck by a sting‐ray barb. But this is exactly what happened to a renowned Australian. While the probability of dying from being stung by a stingray while swimming in the ocean is extremely unlikely, it happened in this instance due to individual factors operating at the time (e.g. his proximity to the sting‐ray, his behaviour, the stingrays behaviour, the location of the barb entering his heart and its removal causing fatal cardiac tamponade).
The imperative to provide an opinion about plausibility (which considers the specifics of the case) as opposed to what is more likely based on general features can be further illustrated in the following example. It is generally agreed that the most likely cause of a fever in any young child presenting to hospital is a viral illness. However, doctors must still consider rare causes of fevers such as meningococcal disease. This requires a shift in thinking from likelihood based on prevalence, to considering the specifics of the particular case, which then drives the need for appropriate investigations and treatment, given the gravity of making an error in this context.
Applying System II Reasoning: An Example
In the case example, the findings together indicate the effects of acceleration/inertial forces to the head reaching (in biomechanical terms) the ‘failure limit’ of the individual bridging vein. The central question is how much force is involved – whether it is necessarily a high force event (and therefore fatal child abuse derived from a system I approach) or whether lower forces may in this instance be responsible (which uses systems II analysis of specific facts in the case). System I thinking invokes the ‘script’ that infants do not die from short falls. This is further reinforced by thinking that if they did, hospitals would be full of infants falling and dying as they learn to stand and climb, which clearly does not happen.
There are, however, certain very rare circumstances under which lower inertial/acceleration forces can be produced inside the skull from external cranial impact and cause subdural haemorrhage (SDH). This statement has arisen from systems II thinking. The recent consensus paper on abusive head trauma acknowledges these very rare events do exist and other papers examining this issue have reached similar conclusions.22, 23 These conditions include children with predisposing conditions such as inherited bleeding disorders and children with rare metabolic brain disorders causing brain atrophy. There is also some evidence that children with expanded subarachnoid spaces, like those with brain atrophy, have an increased baseline stretch on the bridging veins, lowering the force thresholds needed to cause bridging vein rupture in the event of accidental external cranial impact.22, 24-26
The hyperacute rapidly expanding mass‐effect SDH causing rapid rise in intracranial pressure as the cause for the retinal haemorrhage pattern conforms with the mechanism (by Shiau and Levin): ‘There are isolated case reports that severe hyperacute intracranial pressure (ICP) elevation, unlike the sub‐acute pressure increase in abusive head injury, in children may rarely result in extensive retinal hemorrhage’.27 These authors believed such rare cases would be readily identifiable and ‘readily distinguished’ from child abuse. This may not happen in real‐world child protection evaluations if there is sole reliance on system I.
Conclusion
This viewpoint has summarised the dual processing metacognitive models of system I and system II and how to apply this to the evaluation of injury in child protection practice. Both systems have something to contribute but they also carry risk. System I is useful to rely on in the first instance to make immediate decisions about what cases should be investigated but carries the risk of introducing error through cognitive biases. System II is needed at a later stage when the specific facts are defined which considers injury plausibility and contributes to legal decision‐making including within the criminal justice system but also carries risk of error. In the example given those facts are broader than just the height of the fall (a short fall or otherwise), the head injury (SDH and retinal haemorrhage) and the child's death. By shifting from system I initially (to initiate an investigation) to system II, this allows the clinician to avoid giving the incorrect 10 cent answer and instead, with effort, attention and time, provide a more reasoned analytical opinion. In child protection education and in clinical practice we need to be aware of how we think and apply the right system at the right time to arrive at the right responses and conclusions.
No comments:
Post a Comment