This critical analysis will evaluate the clinical decision-making process during an incident that I attended but was led by another paramedic, where there were several possible patient diagnoses and three potential treatment pathways. The probabilities of diagnoses, degrees of certainty and permissible risks will be analysed during the assessment, consideration of diagnoses and treatment stages of this incident. Particular attention will be paid to the influence of emotional biases on our deductive process throughout.
This analysis is anonymised in accordance with the 2018 Data Protection Act and the Health and Care Professions Council Standards of Conduct, Performance and Ethics (2016).
The incident
This incident was assigned to us at 22:25, 35 minutes before our shift was due to finish. We were tired, and our desire to finish on time made us acutely aware of how close this was to the end of our shift.
The patient, who was in his 80s, was sent an ambulance because of an episode of confusion during the evening, which had now resolved. He had a 2-month history of a cough and had had his influenza vaccination 30 hours before this incident. The patient was tachypnoeic, but my impression was that he appeared otherwise well; he was self-mobilising, fully orientated and reported no pain or impairments at that time. The patient's wife, also in her 80s, was in attendance and mobilising using a frame. However, it was apparent that the patient and his wife were dependent on each other for their personal care, with no relatives or friends able to assist them.
On examination, the patient had a peripheral body temperature (PBT) of 38.6°C and a respiration rate (RR) of 26 breaths per minute (bpm). All his other baseline observations were unremarkable and his chest was clear of adventitious sounds on auscultation. We were attending the patient in his home, which was unkempt but warm.
We considered that, alongside his overall presentation, the patient's elevated PBT and RR were primarily indicative of infection, sepsis or an adverse reaction to his influenza vaccination. Without further tests, possible only in definitive care, we could not be certain of the his diagnosis.
At this time, there were three treatment pathways available to us: leave the patient at home with advice on what to do if his condition worsened and recommend that he contact his GP in the morning to discuss his symptoms; remain at the scene and contact an out-of-hours (OOH) doctor to discuss immediate assessment at home; or convey the patient to an emergency department (ED) for assessment and treatment if required, having first instigated care for the patient's wife in the absence of her husband.
We were aware that waiting on the scene for a call from the OOH doctor or to organise care for the patient's wife's at this time of night would result in significant time being spent at the scene. It soon became apparent that we were manipulating the available information and further assessment decisions to conform to our desired belief that the patient's elevated PBT and RR were a self-limiting reaction to his influenza vaccination. If this were so, it would be safe to leave him at home, and we would make a timely finish. This was the incident outcome.
The desire not to finish late caused a significant emotional bias in our decision-making (Lempert and Phelps, 2013). Analysis of this incident will show that this bias caused the adoption of irrationally rapid and therefore potentially erroneous diagnosis forecasting (Blumenthal-Barby and Krieger, 2015).
Our diagnosis forecasting was affirmed by biases of confirmation (manipulating information to conform to a desired outcome or belief) and anchoring (depending too heavily on an initial piece of information) (Shaffer et al, 2016; Featherstone et al, 2020; Stephens, 2020). These human factors (Collen, 2017) were apparent throughout the incident.
Patient assessment
The patient's elevated PBT was indicative of an immune response to illness or infection (Lu and Dai, 2009; Rubia-Rubia et al, 2011; Podsiadło et al, 2019). The origin of his elevated RR could have been metabolic acidosis, indicative of systemic disorders such as infection or sepsis (Cretikos et al, 2008; Rolfe, 2019; Martin-Rodriguez et al, 2020), but that can also occur in response to an increase in blood temperature as a mechanism of physical thermoregulation (Onisor et al, 2019). These factors in the patient's assessment, and his history of a cough and recent episode of confusion were the primary factors in the clinical decision-making in this incident.
Multiple theories explain the process of clinical decision-making, but all these theories share two distinct aspects: the formation of a hypothesis; and the accumulation and deduction of data from the patient's assessment and history, to confirm or reject this hypothesis (Lempert and Phelps, 2013; Monteiro et al, 2018).
We identified the patient's recent influenza vaccine as a plausible hypothesis to explain his elevated PBT and RR. Confirmation and anchoring biases were instrumental in our failure to continue to hypothesise differential diagnoses (Yousaf et al, 2020), thereby avoiding the necessity for the further assessments required to properly inform analysis of these diagnoses.
Further assessments would have included systematic observation, auscultation and palpation of the patient's thorax and abdomen, and rigorous history-taking to enable us to assess the diagnoses' pre-test probabilities (Brown and Cadogan, 2016; Bickley and Szilagyi, 2017). Our failure to carry out these assessments was in contravention of rational reasoning and rigour in our deductive process (Simmons, 2010) as we had insufficient facts upon which to base an informed decision. Therefore, our deductive process was irrational (Blumenthal-Barby and Krieger, 2015).
Benner et al (2016) describe how early formation of a hypothesis becomes an increasingly intuitive process with experience, and that the risk of bias is greatest at this stage of the clinical decision-making process. The deductive stage is analytical and needs to be systematic to mitigate errors in clinical judgement; at this stage, the risk of a disorganised deductive process becomes greater (Blumenthal-Barby and Krieger, 2015).
Probability of diagnosis and differential diagnoses
As we did not consider the pre-test probabilities that the patient's presentation was caused by his influenza vaccine or differential diagnoses of infection or sepsis, we could not have assessed our degrees of certainty, nor considered whether the risks of leaving him at home were permissible.
At the time of this incident (November 2020), older adults (aged 65 years or over) in England were offered the adjuvanted trivalent influenza vaccine (aTIV) (Public Health England, 2020). If the patient's elevated RR and PBT were a result of his aTIV administration, these would have been systemic adverse reactions (Seqirus, 2020). Systemic adverse reactions to vaccinations are classified as serious (Kroger et al, 2022).
Moro et al (2020) searched the Vaccine Adverse Event Reporting System (VAERS) for adverse reactions occurring following the administration of IIV3-HD, an aTIV licensed for use in the United States for people aged over 64 years between 2011 and 2019. Over this period, VAERS received 12 320 reports, of which 723 were serious. Of these serious reports, 30.2% were for pyrexia. However, of the 11 597 non-serious reports, 13.8% were also for pyrexia, meaning that, of the total 12 320 reports, 14.7% (1820) were for pyrexia. All reported cases of pyrexia were self-limiting.
There is a paucity of data to show how long after vaccination pyrexia most commonly occurs. However, Stenger et al's (2020) randomised prospective intervention study of influenza vaccination side effects in athletes reported that the most severe side effect observed was pyrexia and, in their study population, this occurred at 24 hours following administration of the vaccine.
The most likely differential diagnosis was that the patient had an infection; the most likely origins of infection in older adults are the respiratory and urinary tracts (Ginde et al, 2013; Liang, 2016).
Goto et al's (2016) comprehensive cross-sectional analysis of more than three million older adults admitted to EDs in the US between 2011 and 2012 showed that 13.5% of all admissions within this group were for infection, of which 50.3% were for respiratory tract infections (RTIs) and 25.3% for urinary tract infections (UTIs). Goto et al's (2016) analysis has significant power, rendering the data gathered reliable and reproducible so it is sufficient to inform clinical decision-making.
Both Goto et al's (2016) and Moro et al's (2020) are large-scale studies based in the United States. England is demographically similar to the United States and has a similarly well-funded and modern healthcare system (Keane et al, 2020); therefore, being carried out in different countries should not be considered a barrier to comparison.
RTIs and UTIs are the most common precursors to sepsis in older adults (Martin et al, 2006), who are significantly more likely to develop sepsis if the source of the infection is bacterial; of all infections, around one-third are bacterial and can therefore be treated with antibiotics (Liang, 2016).
Martin et al's (2006) longitudinal observational study of 10 422 301 adult sepsis patients between 1979 and 2002, with its sample size giving it significant power to inform, showed that 37.1% of sepsis cases originated in RTIs and 28.2% originated in UTIs.
Ginde et al's (2013) retrospective analysis of the impact of older age, comparing patients from their own home with those from residential care, found that 6.5% of older adults admitted to an ED with an infection were diagnosed with sepsis. However, changing the data set classification to compare patients living at home (like the patient in this case) with those in residential care (where pathogen levels are inherently higher), showed that 1.9% of patients admitted from their own home were diagnosed with sepsis, compared with 14% of those admitted from residential care.
Synthesis of the data gathered during the incident in this study, and from analyses by Martin et al (2006), Ginde et al (2013), Goto et al (2016), Moro et al (2020) and Stenger et al (2020) provides enough information to consider the pre-test probabilities of the patient's probable diagnoses (Box 1).
| Pre-test probability calculation summary |
|
|
| Probabililty percentage as number of patients |
|
|
Moro et al (2020) state there is a 14.7% probability of developing pyrexia following aTIV administration and it is self-limiting. Stenger et al (2020) suggest that the onset of pyrexia is likely to occur around 24 hours following administration. The patient's episode of confusion, which is commonly associated with the onset of pyrexia (Onisor et al, 2019), occurred at approximately 24 hours, suggesting this was at the onset time of his pyrexia.
RR commonly rises with PBT as the body attempts to regulate metabolic acidosis and maintain thermoregulation (Onisor et al, 2019). Onisor et al's (2019) detailed and methodical two-year clinical study of 362 patients with pyrexia suggests that RR rises by 7–11 bpm per degree Celsius above normothermic. Although there is no consensus on the definition of normothermic in adults, it is widely accepted that 1°C above normothermic indicates pyrexia (Sund-Levander and Grodzinsky, 2009; Niven et al, 2015); if this is applied to the patient, his RR of 26 bpm suggests his PBT was approximately 1°C above the normothermic.
If the patient is considered as one of Goto et al's (2016) 13.5% ED admissions for infection, then there is a 50.3% chance this was for a RTI, of which the patient's history of a cough was also suggestive (Mahashur, 2015); applying the data from Martin et al's (2006) observational study showing that potentially treatable bacterial infections account for one-third of all infections in older adults, of which 37.1% are RTIs that cause sepsis, results in a pre-test probability of 0.84% that the patient had a bacterial RTI that could cause sepsis.
This probability calculation validates Ginde et al's (2013) study showing that 1.9% of older adults admitted to the ED from their own home with an infection diagnosis have sepsis, with half of cases (0.95%) originating from RTIs.
Therefore, it is reasonable to surmise pre-test probabilities of: 14.7% that the patient's presentation was a self-limiting reaction to his influenza vaccine (Moro et al, 2020); 4.5% (one-third of 13.5%) that he had a bacterial infection treatable with a course of antibiotics (Liang, 2016); and 1.9% that it was a precursor to infection that might forebode sepsis requiring hospital admission (Ginde et al, 2013; Goto et al, 2016).
Management and treatment
The National Institute for Health and Care Excellence sepsis screening tool (2017) shows that the patient's RR of 26 bpm placed him at a high risk of sepsis, requiring urgent conveyance to an emergency care setting with resuscitation facilities. The patient's RR and pyrexia gave him a National Early Warning Score 2 (NEWS2) of 4, indicating a potential need for escalation of care (Royal College of Physicians, 2017). Had we followed these algorithm-based screening tools alone, we would have conveyed the patient to an ED.
Algorithm-based screening tools can lower the risk of serious adverse events by improving clinical deterioration prediction (Martín-Rodríguez et al, 2020). However, the accuracy of screening tools can vary considerably.
Smyth et al's (2016) systematic review aimed to determine whether sepsis screening tools were accurate enough to improve clinical prediction; in their review, nine studies including 147 695 patients were analysed, producing data showing that sepsis screening tools have a mean sensitivity of 65.29% and mean specificity of 68.70%.
Martín-Rodríguez et al's (2020) well-designed, multicentre cohort study of 2335 participants, selected through a methodical inclusion criterion, evidenced the predictive accuracy of NEWS2 to have a sensitivity of 70.62% and a specificity of 82.9%.
These data sets both suggest that screening tools are less accurate at identifying those with disease than at identifying those without disease, meaning there is a high chance of patients being conveyed to ED needlessly, potentially exposing them to unnecessary risk (van den Broek et al, 2020). This was particularly relevant to this incident as it occurred during the COVID-19 pandemic, so the risk:benefit ratio of conveying the patient to an ED was greater than usual.
Furthermore, screening tools can only be as accurate as the data used to populate them (Martín-Rodríguez et al, 2020). The patient's NEWS2 score of 4 was owing, in part, to his PBT of 38.6°C. However, Niven et al's (2015) comprehensive systematic review and meta-analysis of 75 studies, with a combined sample of 8682 participants, showed that peripheral thermometers have a low sensitivity of 64%, meaning that our ability to accurately measure the patient's PBT was insufficient to inform our clinical decisions (Niven et al, 2015).
The worst-case scenario was that the patient had an infection that, without antibiotic treatment, may have become systemic overnight. However, this analysis suggests that the risks of leaving the patient at home with his wife, who would have been able to call for an ambulance if his condition deteriorated, were permissible.
Moreover, the patient evidently had a reductionist view of health, meaning he viewed good health as the demonstrative absence of disease (Engel 1981; Tamm, 1993); therefore, as he felt well, he did not consider conveyance to hospital necessary. Finally, it was apparent that separating the patient and his wife would cause them both undue anxiety. To have ignored these facts and provided algorithm-based, impersonal care would have dehumanised the patient (Todres et al, 2009), and neglected his best interests by delivering care that was not patient-centred.
In 2017, the Association of Ambulance Chief Executives (2017) announced its commitment to ‘making every contact count’ by improving the health and wellbeing of those most vulnerable in the community.
Although human factors led us to decide that the patient was safe to be left at home, we were concerned with the unkempt condition of his home. Therefore, even though we were keen to make a timely finish to our shift, we took the time o discuss the patient's ongoing cough, advising that this could be an upper airway cough syndrome, a viral condition best monitored by his GP (Mahashur, 2015), and that poor air quality in an unkempt home would aggravate this.
In addition, we considered that the condition of the patient's home could be a result of an increasing inability to adequately cope with caring for his wife, alongside her decreasing mobility, while managing their daily living activities. We believed that a review of the patient's and his wife's care needs might reduce the need for ongoing healthcare interventions, so gained their consent to make a safeguarding referral.
The paramedic who led this incident was experienced, so intuition would have played a meaningful role in their clinical decision-making, drawing upon the sum of their experience to recognise patterns and use tacit knowledge (Benner et al, 2016). Analysis of this incident and of the pre-test probabilities of diagnoses suggests that the paramedic was unconsciously competent, able to recognise patterns and quickly make evaluations based on little information (Bate et al, 2012) without apparent adherence to algorithm-based rules.
In prehospital care, clinicians frequently have to make quick, autonomous decisions based on normative imperfect information (Sherbino et al, 2012; Collen, 2017). Therefore, rapid deduction must not be assumed to be detrimental to the clinical decision-making process, as it can expedite the type of optimal decision that is often necessary in these circumstances (Gigerenzer, 2007). This implies that intuitive decision-making in rapid diagnosis can be more accurate (Sherbino et al, 2012).
Conclusion
Beginning this analysis, I surmised that, because of the influence of biases, we had potentially made the wrong decision to leave the patient at home, and that we should have contacted an OOH doctor or conveyed the patient to an ED. However, I conclude that, in all probability, leaving the patient at home was the correct treatment pathway.
A more in-depth patient assessment would have rationalised our deductive process. However, the evidence presented in this analysis allows for the hypothesis that the paramedic lead in this incident acted upon a sound intuition and tacit knowledge that the patient was safe to be left at home.
Aristotle (2014) described bias as a deviation from the mean towards an extreme. On the balance of the probability of diagnoses, risk:benefit ratios and permissible risks, leaving the patient at home was the mean point and, perhaps, no amount of bias would have influenced an alternative pathway against an intuition that it was unsafe or not in his best interests.