Influence of Genetic Information on Neonatologists’ Decisions: A Psychological Experiment

In the fall of 2020, we emailed a 22-question questionnaire entitled “Making Difficult Decisions in the Neonatal Intensive Care Unit” to roughly 3600 neonatologists. The questionnaire was distributed by e-mail through the listservs of the American Academy of Pediatrics Section on Neonatal and Perinatal Medicine and the Children’s Hospital Neonatal Consortium. The latter was included to increase sampling of those working in level IV NICUs in which genetic testing is most extensively used. In accordance with the tailored design method and listserv regulations, we sent 1 reminder to the former group and 2 to the latter group and offered a $10 Amazon gift card to all who completed the questionnaire.¹³  Study data were collected and managed by using Research Electronic Data Capture electronic data capture tools hosted at the Children’s Hospital of Philadelphia.¹⁴  This study was reviewed and deemed exempt by the Children’s Hospital of Philadelphia Institutional Review Board.

We developed a case-based psychological experiment after review of the literature and consultation with geneticists and neonatologists at level III and IV NICUs. The 4 fictionalized case vignettes described acutely ill infants. We then pilot-tested the cases for clarity with 10 experienced neonatologists and revised them on the basis of feedback. Case 1 is included as an example in Fig 1, and the full questionnaire is available as Supplemental Information. Hypotheses were tested by using a randomized split-block design with alternative versions of the case vignettes, a method used successfully in other studies.^15,16  Each fictionalized case had 2 versions that varied only by the genetic information provided, henceforth referred to as the genetic and nongenetic versions. Participants were randomly assigned to see only one version of each case; they were randomly assigned independently for each case, so a participant might be presented with any combination of the 4 vignettes. Participants were subsequently asked questions about their recommendations for the patient. All responses were recorded on a 5-point Likert scale.

Case 1 described a patient with severe chronic lung disease and asked about provision of tracheostomy and gastrostomy tubes. In the genetic version, the patient had a VUS in a surfactant protein; in the nongenetic version, she had negative genetic testing results. Case 2 described a preterm infant with sepsis and asked about provision of a central line to continue antibiotic therapy. In the genetic version, the patient was incidentally diagnosed with Williams syndrome; in the nongenetic version, he had negative genetic testing results. Williams syndrome is a gene deletion associated with intellectual disability (typically mild) as well as unique personality characteristics and endocrine and cardiac abnormalities.¹⁷  Case 3 described a patient with pulmonary hypoplasia and participants were asked about extracorporeal membrane oxygenation (ECMO) candidacy. In the genetic version, the hypoplasia resulted from a mutation with a broad spectrum of outcomes; in the nongenetic version, the etiology was oligohydramnios. For case 3, the following prognostic statement was included after both versions to mitigate uncertainty about disease severity: “Expert consult suggests that outcomes may range from a severe asthma-like syndrome to lethal pulmonary hypoplasia, but prognosis is difficult to assess.” After the first 3 cases, participants were asked whether they would recommend the stated invasive intervention and whether they would recommend palliative care. They were also asked to rate the importance of each of the following factors in making their recommendations: likelihood of short-term survival, likelihood of long-term survival, pain and suffering for the patient, expected long-term neurologic outcome, effect on the child’s family, and allocation of limited resources.

In case 4, the patient was a term male with progressive muscle weakness and ventilator dependence. The genetic version included pending whole exome sequencing, whereas the nongenetic version had no mention of genetic testing. Participants were asked about readiness for a goals of care conversation. T hey were also asked to what degree the following information would inform their recommendations for goals of care: need for long-term intravenous nutrition, brain MRI results, social work assessment of caretakers’ ability to manage a child requiring complex medical care, otolaryngology consult regarding candidacy for tracheostomy, and whether a genetic etiology was found. Finally, respondents completed additional questions about demographic details and work environment.

Responses were analyzed by using Stata version 16.1 (StataCorp, College Station, TX). Statistical inferences were based on two-tailed tests, with significance set at P < .05. Descriptive statistics for physician characteristics and practice settings are presented for the full sample as well as for randomization group by case and are reported as number of participants and percentages. The recommendations for invasive and palliative interventions served as our primary outcome for each vignette. Because this outcome was ordinal output from a 5-point Likert scale, we used the Wilcoxon–Mann-Whitney test to assess for differences in responses between the groups that saw genetic and nongenetic versions of each case. We calculated Cohen’s d effect size for outcomes in which the genetic and nongenetic group had significantly different responses because this is a commonly used measure of the importance of an effect in decision-making.^18,19  In a secondary analysis, we used ordered logistic regression to correct for any failures of randomization that had led to unequal distribution of respondent characteristics between the genetic and nongenetic groups.

A total of 551 respondents completed the entire questionnaire. This represents a response rate of 40% (18 of 45) for the Children’s Hospital Neonatal Consortium group, 21% (533 of 2550) for the American Academy of Pediatrics group, and 21% for the cohort overall. The characteristics of respondents are shown in Table 1.

Figure 2 displays responses for the primary outcome questions from the 4 clinical cases. In case 1, the finding of a VUS was associated with participants being less likely to recommend a tracheostomy and gastrostomy tube (P < .001; Cohen’s d: 0.55) and more likely to recommend transitioning to palliative care (P < .001; Cohen’s d: 0.46). This confirmed our hypothesis that presence of a VUS would be associated with neonatologists being less likely to recommend invasive care and more likely to recommend palliative care. Pain and suffering were ranked as more important decision-making factors for those who saw the case with the VUS (P = .01).

In case 2, an incidental finding of Williams syndrome was associated with participants being less likely to recommend central line placement (P < .001; Cohen’s d: 0.35) and more likely to recommend transitioning to palliative care (P < .001; Cohen’s d: 0.37). This confirms that including an incidental genetic finding with neurodevelopmental implications (even mild ones) would be associated with favoring palliative over invasive interventions. Here, pain and suffering (P = .01) and long-term survival (P = .03) were cited as more important factors in the recommendations of participants who saw the version with Williams syndrome.

In case 3, providers were more likely to recommend transitioning to palliative care if the patient had a genetic etiology of disease (P = .004; Cohen’s d: 0.24). There was no significant difference in recommendations about ECMO between the genetic and nongenetic groups (P = .09). This partially supported our hypothesis in that a genetic etiology of disease meant providers were more likely to recommend palliative care and less likely to recommend ECMO. Rationales for decisions did not differ significantly between groups for case 3. Secondary analysis to account for an imbalance in years in practice in case 3 had a minimal effect on the results.

In the final case, reported readiness for a goals of care meeting did not differ significantly if whole exome sequencing was pending (P = .06). Providers ranked finding a genetic etiology as the most important factor informing their recommendations to the family (mean: 2.75). MRI results were second most important in informing recommendations (mean: 2.63), followed by whether the patient could tolerate oral feeds (mean: 2.47), social work assessment (mean: 2.34), and tracheostomy candidacy (mean: 2.17).

We used an experimental design to examine how genetic information affects neonatologists’ clinical decisions regarding case vignettes. To capture the inherent complexity and ambiguity associated with genetic test results, we chose results that either have uncertain prognostic value or are associated with a range of clinical outcomes. We studied the impact of 4 types of genetic results on care recommendations: (1) a VUS; (2) a genetic diagnosis with implications for neurodevelopment; (3) a genetic versus nongenetic etiology of disease; and (4) pending genetic testing results. Overall, the presence of genetic findings was associated with providers being less likely to recommend invasive intervention and more likely to recommend a transition to palliative care in 3 of 4 cases.

In the first case, the influence of a VUS on care is, by definition, not supported by prognostically informative evidence. Clinical geneticists agree that VUSs should not be used in medical decision-making.²⁰  This constitutes experimental evidence that ambiguous genetic information inappropriately influences how providers reach clinical decisions. For this case, respondents cited pain and suffering as more important considerations if they saw the genetic version of the case and were thus more likely to recommend palliative care. There is no reason to believe that a VUS would be associated with increased pain and suffering. This pattern is in line with previous work documenting that physicians cite quality of life more often when recommending palliative, rather than invasive, interventions.²¹  It is also consistent with reported use of suffering to represent the general utility of an intervention and justify a wide variety of decisions.^21,22  The reason that a VUS biased respondents to recommend palliative care may be that they suspected this genetic finding would worsen prognosis, compounded by desperation for information when faced with a difficult clinical decision.

In case 2, we selected Williams syndrome to isolate the effect of intellectual disability from survival likelihood. This diagnosis is associated with intellectual disability (usually mild) but should not affect likelihood of acute survival in the absence of structural anomalies. The altered responses in the genetic version of this case reveal that genetic testing heralding future disabilities changes care decisions. Because we did not directly evaluate whether participants thought Williams syndrome affected survival, inaccurate presumptions about survivability may have informed recommendations in addition to predicted intellectual disability. In case 2, pain and suffering as well as long-term survival were cited as a more significant factor for those considering a patient with an incidental Williams syndrome diagnosis. Current life expectancy estimates are lacking for patients with Williams syndrome, yet survival well into older adulthood in most cases makes this a questionable rationale. The effect of Williams syndrome on care recommendations may have been due to misinformation about prognosis or resulted from generalized concerns about quality of life for a patient with intellectual disability.

In case 3, a genetic etiology of disease was associated with an increased likelihood of recommending palliative care despite seeing the same direct prognostic statement in both versions of the vignette. This suggests that having a genetic diagnosis is perceived as an inherent marker of a bad outcome even if available evidence does not support a change in prognosis. This further supports the overall conclusion that genetic findings foster generalized pessimism for neonatologists facing difficult decisions.

In the final case, pending whole exome sequencing was not related to perceived readiness for a goals of care meeting. However, a genetic etiology of disease was regarded as the most important factor informing goals of care, above factors with clearer implications for management needs such as tracheostomy candidacy. This reinforces the general theme of our study: genetic information affects how neonatologists think about difficult care decisions. Effect sizes throughout our study also support this conclusion. Effect sizes are most meaningfully compared with estimates from other similar studies; however, there are no comparable data in this case. Instead, we compare our effect sizes to reference values. They are in the moderate range, at 0.3–0.5, showing that changes in genetic information have a substantial effect on neonatologists’ decisions.

Our study design, a randomized controlled psychological experiment, creates both strengths and limitations. Randomization and control enhance internal validity. In each scenario, we were able to isolate one aspect of genetic information (the independent variable) and evaluate the effect of experimentally controlled variations of that aspect on neonatologists’ care recommendations (the dependent variable), enabling us to conclude that the observed effects were due to our manipulation of aspects of the genetic information.^23–25  Doing anything remotely similar to this in a direct clinical study of patients would be not only difficult but, likely, also unethical. An additional strength of our design was that we oversampled neonatologists who work at level IV NICUs, 61% of our sample. Thus, we capture a population of neonatologists who most often order genetic tests and use the results to inform clinical decisions.

Our study design also has limitations. First, we do not know the mechanism underlying the observed results. Genetic findings could affect clinicians’ thinking about care decisions (in practice or in response to this study) in at least two ways: because of the specific genetic content (which presumably is the major influence) or because a genetic finding causes information overload by adding to the overall complexity of the case. Information overload has been widely recognized to impede clear medical decision-making and may lead clinicians to perceive illness as more severe.²⁶  Second, we cannot determine the extent to which the physicians’ responses in this experiment reflect their actions in clinical practice.¹²  Substantial evidence exists, however, that clinical vignettes accurately capture how physicians practice.^15,16  Third, the response rate to the current study, slightly >20%, may limit the generalizability of our results. Although this response rate exceeds the typical response rate of 10% to 15% reported for the Section on Neonatal and Perinatal Medicine listserv, our findings may not generalize to neonatologists who declined to participate.²⁷  Notably, this generalization failure would most likely occur if respondents were both more likely to participate and less likely to use genetic information appropriately, but this seems unlikely, particularly given the high percentage of respondents who work at level IV NICUs. Nonetheless, because this is the first study to look at biases in neonatologists’ decisions as they relate to genetic information, true external validity will arise only from replication of such findings in different contexts and by using different methodologies.²⁸ 

If the biases demonstrated in our study extend to clinical practice, they have a number of troubling implications. Our findings suggest that ambiguous genetic information may be more than unhelpful; it may be detrimental to clear and rational decision-making by physicians. Also troubling, diagnoses portending intellectual disability may be used to redirect care. If such use occurs in practice, it sends a hurtful message about the value of disabled persons. Disability advocates have also raised concerns that using genetic findings in this way may stem from misunderstandings about what it is like to live with or care for someone with a disability.⁵ 

We found that, in responding to clinical vignettes, neonatologists make treatment decisions that are influenced by genetic information. As the prevalence of genetic testing expands in the NICU, recognizing biases associated with genetic diagnoses and facilitating appropriate use of genetic findings by neonatologists, in collaboration with geneticists, is paramount. We hope that these results may raise neonatologists’ awareness of their biases with regard to both genomic information and intellectual disability. Future work should evaluate the effectiveness of educating clinicians on interpretation of genetic information and creating decision-making frameworks to support neonatologists’ use of this ever-expanding and inherently complex information.

We acknowledge Erik Brandsma and Karen Puopolo, who both provided thoughtful feedback on early drafts of our clinical vignettes.

ECMO

extracorporeal membrane oxygenation

VUS

variant of uncertain significance