The relative lethal toxicity of pharmaceutical and illicit substances: A 16-year study of the Greater Newcastle Hunter Area, Australia
Abstract
Aims
We aim to calculate 2 metrics of relative lethal toxicity; the fatal toxicity index (FTI; number of deaths per year of a daily dose) and the case fatality (CF; number of deaths per overdose) with a focus on opioids, antidepressants, antipsychotics, benzodiazepines and illicit drugs.
Methods
This descriptive cohort study used the Australian National Coronial Information System (NCIS) to identify a population of individuals with drug-associated deaths in the Greater Newcastle Hunter Area between January 2002 and December 2016. This was combined with Australian medicine dispensing data and corresponding data from the Hunter Area Toxicology Service to calculate FTI and CF.
Results
There were 444 drug-related deaths and 21,296 overdoses during the study period. FTI and CF were well correlated (Spearman's rho 0.64, P < .001). Of the classes of interest, opioids had the highest FTI (40.3 95% confidence interval [CI] 35.2–45.4 deaths per 100 years of use at the defined daily dose or deaths/DDD/100 years) and CF (12.4% 95%CI 11.0–13.9). Fentanyl, methadone and morphine had the highest relative fatal toxicity within this class. Tricyclic antidepressants had the highest relative fatal toxicity of all antidepressants (FTI 14.5 95%CI 9.7–19.3 deaths/DDD/100 years and CF 7.1% [95%CI 4.8–9.3]) and benzodiazepines appeared to be more associated with multiple agent deaths than single. Of the illicit drugs, heroin had the highest CF (26.4%, 95%CI 19.1–33.7).
Conclusion
Knowledge of relative lethal toxicity is useful to prescribers and medicines and public health policy makers in restricting access to more toxic drugs and may also assist coroners in determining cause of death.
What is already known about this subject
- The relative lethal toxicity of drugs can be estimated according to the fatal toxicity index (FTI; the number of deaths per year of a daily dose) and case fatality (CF; the number of deaths per overdose).
- Challenges in access to population level prescribing, poisoning and death data have hindered past attempts to estimate the relative lethal toxicity of drugs.
- This study estimates the relative lethal toxicity of drugs taken in overdose, which is of direct relevance to prescribers and policy makers.
What this study adds
- Population level death, prescribing and overdose data in Australia can be used to calculate FTI and CF for poisoning with pharmaceutical and illicit agents.
- FTI and CF calculated within this study are relatively well correlated.
- As a class, opioids had the highest FTI and CF, tricyclic agents were the most toxic antidepressants and benzodiazepines were significantly more toxic in combination overdoses compared to when taken alone.
1 INTRODUCTION
Death by poisoning is a major international concern and most often occurs in the context of deliberate overdose. Toxicovigilance is a term recently used to describe the practice of monitoring for the emergence of poisons causing significant morbidity and mortality in overdose.1 For pharmaceutical agents, this field of research faces similar difficulties to pharmacovigilance in that it occurs exclusively in a postmarketing environment that lacks the structure, regulation and economic incentives of the premarketing setting.2 For patients at risk of overdose, knowledge of the relative lethal toxicity (e.g. the risk of death following a prescription or overdose relative to other medicines) is important as it may influence the decision to prescribe and choice of medicine prescribed. At a regulatory and public health level, this information can generate safety signals and inform policy decisions.3-5
Estimating the relative lethal toxicity of poisons poses several challenges. Animal toxicity studies (e.g. LD50 data) cannot be reliably extrapolated to humans6, 7 and so information regarding human toxicity is frequently derived from case reports and case series of single drug overdoses. However, these sources are subject to significant publication bias and only occasionally include patients who die as a result of their poisoning.
Case fatality (CF) is a metric of relative toxicity that estimates of the risk of death following overdose with a given agent. It is calculated by expressing the number of deaths associated with a given poisoning as a percentage of the number of people presenting to hospital with that poisoning. There have been large poisoning cohort studies estimating in-hospital case fatality.6 However, this approach misses deaths from poisonings in which individuals die before reaching hospital, which represents the majority of people who die from many highly toxic agents and so, in most cases, grossly underestimates true case fatality.
Another approach is to calculate the fatality toxicity index (FTI); an estimation of the risk of death after a given agent is prescribed. This metric has typically been expressed as the number of deaths per million prescriptions or per million patient-years and has been applied to antidepressants,8 benzodiazepines9, 10 and antipsychotics.11 However, no studies to date have systematically evaluated the fatal toxicity of the full range of agents involved in poisoning deaths. As a measure of lethality, the FTI has limitations including those introduced by the use of dispensing claims data and bias related to prescribing based on risk of overdose.11, 12 Furthermore, FTI cannot be used to quantify the relative lethality of nonprescription pharmaceuticals and nonpharmaceutical agents such as illicit substances.
In this study, we aimed to determine relative lethal toxicities of poisons causing deaths reported to the coroner within the Greater Newcastle Hunter Area over a 16-year period. Specifically, we calculated the CF and FTI for individual agents and classes of agents to identify those with relatively higher lethal toxicity. We also evaluated the correlation between CF and FTI.
2 METHODS
2.1 Study populations
2.1.1 Drug-poisoning related deaths
We identified drug-poisoning related deaths (pharmaceutical and illicit) in the Hunter Valley Region, New South Wales, Australia between January 2001 and December 2016 using the National Coronial Information System (NCIS) database. The NCIS was established in July 2000 and is a national repository of coded information on all deaths referred to the Australian coroner, including all poisoning deaths. It contains a unique person identifier, date and geographic location of death, case status, and information on mechanism and cause of death. Each death record may also contain autopsy, police and forensic reports.
We extracted all closed coronial deaths occurring within statistical areas 4 (SA4) 106 (Hunter Valley excluding Newcastle) and 111 (Newcastle and Lake Macquarie) within the NCIS database between 1 January 2001 and 31 December 2016 inclusive, and then limited our search to those where the primary mechanism of death was coded as exposure to chemical or other substance. We searched the 4 primary cause of death fields (Cause of Death 1a to d) to identify all deaths where pharmaceutical or illicit drugs were listed. We then attempted to identify all drugs involved with that death by manually reviewing all other fields within the NCIS dataset.
We manually identified misspellings of drugs and converted brand to generic drug names to ensure complete data capture (Supplementary S1). In anticipation of causation being attributed to metabolites, we recoded drugs that were likely to be metabolites as the parent compound (as identified by clinical pharmacologist authors J.B. and N.B.). For example, if amitriptyline and nortriptyline were both listed as associated with a death, nortriptyline was assumed to be a metabolite and the death was attributed to an amitriptyline overdose only (Supplementary S2). For death records in which only a drug class was listed or nonspecific terminology was use (such as drug), we evaluated autopsy, forensic and police reports to identify the drug(s) most likely to be associated with death. A drug-poisoning related death was included if at least 1 drug could be identified down to at least the drug class level.
For our primary analysis of deaths involving multiple drugs, we anticipated that some drugs would contribute more to the death than others, therefore we weighted the drug-poisoning related deaths according to the number of drugs implicated in each death. For each death, each attributed drug was therefore assigned a value equivalent to the reciprocal of the total number of agents associated with that death. For instance, if 6 drugs were involved in a death, each drug was considered to be contributing to 1/6 of a death. All of these weighted deaths were summed for each drug to calculate the overall weighted number of deaths per drug.
2.1.2 Hospital presentations with drug overdose
We identified all drug overdose-related presentations recorded within the Hunter Area Toxicology Service (HATS) database for the study period. The HATS database was established in 1987, and represents 1 of the largest poisoning cohorts in the world.13 It provides information on all poisonings presenting to hospitals within a well-defined geographic region and contains a unique person identifier, date of presentation, reason for admission, poison(s) involved and whether the overdose resulted in death. We included admissions that were due to a drug overdose, excluding those with concurrent other reasons that may have superseded the drug as a reason for presentation. Such examples included exposure to pesticides and other toxic chemicals, carbon monoxide and other medical diagnoses or physical self-harm attempts, with the exception of alcohol intoxication and wrist/arm lacerations. We performed sensitivity analyses including all of these excluded presentations to evaluate the impact on our estimates of relative lethal toxicity.
As for the drug-poisoning related deaths, we weighted the hospital presentations with drug overdose according to the number of implicated agents in each overdose.
We correlated HATS deaths to deaths recorded in the NCIS database and reviewed cases in which a death was recorded within our HATS study population but not within our NCIS study population to determine why this may have occurred.
2.1.3 Pharmaceutical drug prescriptions
The Australian Statistics on Medicines (ASM) dataset is an annual publication produced by the Drug Utilisation Sub-Committee of the Pharmaceutical Benefits Advisory Committee and provides medicine utilisation data in the form of defined daily doses per 1000 population per day (DDD/1000/day) for all medicines funded by the Australian government's universal health care programme; the Pharmaceutical Benefits Scheme. The DDD is a World Health Organization approved measure of drug utilization and represents the assumed daily dose of the drug when used for its main indication.14
2.1.4 Population estimates
We used Australian Bureau of Statistics (ABS) mid-year population estimates for SA4 106 (Hunter Valley excluding Newcastle) and 111 (Newcastle and Lake Macquarie) as population level denominators, taking the average population size over the 16-year study period as the population at risk.
2.2 Data analysis
Our primary analysis included all drug related deaths, and our secondary analyses considered deaths only involving a single drug. Data coding and manipulation and statistical analysis was performed in Microsoft Excel 2016, SAS/STAT version SAS/STAT version 13.2 of the SAS system (version 9.4) for Windows and Stata v15.1 (StataCorp, Texas, USA).
2.2.1 CF
We calculated the number of overdoses with each drug by adding the number of overdoses from the HATS study population to the number of drug-poisoning related deaths from the NCIS study population (to account for overdoses leading to out of hospital deaths). We then subtracted the number of deaths from the HATS study population to avoid double counting.
CF was calculated by dividing the number of drug-poisoning related deaths by the number of hospital presentations with overdose for a given drug, expressed as a percentage with 95% confidence intervals (CIs). CIs for each drug were calculated using a binomial distribution in which the number of deaths associated with each drug represented the number of events and the number of overdoses represented the number of trials. For primary analyses, weights were applied to deaths and overdoses as above.
2.2.2 FTI
We calculated FTI as the number of deaths per 100 years exposed to the DDD of that drug. For each medicine we used the number of deaths associated with that medicine as the numerator. For the denominator of each drug, we summed the DDD/1000pop/year over the entire 16-year study period and multiplied by the population size in in Greater Newcastle Hunter area; the average ABS mid-year population for SA 106 and 111 combined over the study period. We then divided this by 10 to give a metric of deaths per 100 years of use at the DDD (deaths/DDD/100 years). The value of DDD/100 years value has little meaning in isolation but is used to determine the FTI of substances relative to each other. A binomial distribution was used to calculate 95% CIs around FTI, assuming the number of deaths was the number of events and DDD/year is the number of trials. Deaths were weighted in the primary analysis as above.
2.2.3 Drug class analysis
We focused our analyses on specific drug classes frequently taken in overdose and anticipated to be frequently associated with drug related deaths; these included opioid, antidepressant, antipsychotic and benzodiazepine drug classes as defined by the Australian Medicines Handbook.15
In our primary analysis, we calculated CF and FTI for these classes for all deaths and then separately for deaths only involving single agents. For the situation in which a death or overdose included more than 1 drug of the same class, we considered these to be unique drugs when calculating weighted deaths and overdoses.
We calculated CF and FTI for all medicines taken in overdose and within the HATS study population, regardless of whether they resulted in a death and used these values to calculate median values of CF and FTI for all drugs and classes considered in our primary analysis, and only drugs involved in single drug deaths in our secondary analysis to facilitate observations of relative lethal toxicity. We retained methadone, buprenorphine, codeine and pholcodine within these analyses, despite significant limitations related to capture of dispensings (see limitations).
2.2.4 Data visualisation and correlation analyses
Only drugs that were associated with a death were presented in this analysis. To summarise both the FTI and CF for each drug class of interest as well as the correlation between FTI and CF measures, we plotted the weighted FTI against the weighted CF with CIs for each drug within each class of interest. We also presented an extrapolated line representing the median weighted FTI and CF for that drug class to illustrate the point of correlation between the 2 measures. For illicit drugs, only weighted CF was plotted.
Where values for both FTI and CF existed for drug deaths, we used Spearman rank correlation techniques to quantify their correlation, where P < .05 was considered significant.
To compare our primary (all drug deaths) and secondary (single drug deaths) analyses we plotted weighted FTI and CF against FTI and CI calculated for single drug deaths for each drug class of interest respectively. We used this to identify agents that had a substantially different lethal toxicity in single vs multiple agent overdoses.
Hunter New England Human Research Ethics Committee approved the use of HATS data (15/02/18/5.04) and Victorian Department of Justice Research Ethics Committee (CF/15/18367) approved the use of NCIS data. ABS and ASM data are publicly available.
3 RESULTS
3.1 Drug-poisoning related deaths
We identified 444 drug-poisoning related deaths within the Hunter Valley Region between January 2001 and December 2016 using the NCIS dataset (Figure 1). Most decedents were male (n = 279, 62.8%) and the median age of death was 43 years (range 1–94 years; Table 1). Deaths mostly occurred at home (n = 332, 74.8%), or in hospital 73 (16.4%,). Most deaths were associated with pharmaceutical agents only (n = 342, 77.0%), 55 (12.4%) with an illicit agent only and 47 (10.6%) with a combination of pharmaceutical and illicit agents. A total of 213 (48.0%) deaths were associated with a single drug. There was a total of 99 unique specified drugs associated with these deaths: 89 pharmaceutical and 10 illicit agents. The median number of agents causing death was 2 (range 1–11).
| NCIS demographics (n = 444 deaths) | HATS demographics (n = 11 907 admissions) | |
|---|---|---|
| n (%) | n (%) | |
| Gender | ||
| Male | 279 (62.8) | 4211 (35.4) |
| Female | 165 (37.2) | 7671 (64.4) |
| Transgender | 0 | 25 (0.2) |
| Age (y) | ||
| ≤25 | 31 (7) | 3516 (29.5) |
| >25 to ≤35 | 96 (21.6) | 2773 (23.3) |
| >35 to ≤45 | 123 (27.7) | 2662 (22.4) |
| >45 to ≤55 | 125 (28.2) | 1793 (15.1) |
| >55 | 69 (15.5) | 1163 (9.8) |
| Death location | ||
| At home | 332 (74.8) | NA |
| In hospital | 73 (16.4) | NA |
| Other | 39 (8.8) | NA |
3.2 Hospital presentations with drug overdose
We included 11 907 presentations with overdose relating to 21 296 drug overdoses to HATS (Figure 2). Most people overdosing were female (n = 7671, 64.4%) and the median age was 34 years (range < 1–99 years; Table 1). Most overdoses were associated with a pharmaceutical agent only (n = 11 208, 94.1%). There were 408 (3.4%) overdoses associated with an illicit agent only and 291 (2.4%) with a combination of pharmaceutical and illicit agents. There were a total of 6632 (55.7%) overdoses with a single drug. HATS recorded drugs differently to NCIS, including combination products and varying dosages of certain products, precluding direct comparison; however, there was a total of 509 listed drug components associated with the overdoses: 490 pharmaceutical and 19 illicit agents. The median number of agents in overdose was 1 (range 1–14). Within the 11,907 admissions, there were 24 deaths. Eight of these deaths did not appear within the NCIS study population: 4 deaths were presumptively matched in NCIS, while 4 could not be identified and were presumably not referred to or investigated by the Coroner.
3.3 FTI and CF
For our primary analyses, the median FTI was 5.38 deaths/DDD/100 years and CF was 4.26%. The weighted FTIs and CFs for individual products and for drug classes can be found in supplementary information (Supplementary 3). The correlation between weighted FTI and CF for all 81 individual drugs implicated in a drug death where both FTI and CF was available was strong (Spearman's rho 0.64, P < .001), and very strong for the single drug deaths in our secondary analysis (Spearman's rho 0.96, P < .001). Sensitivity analyses in which we excluded overdoses with drugs in which the presentation to hospital was not primarily with drug toxicity had minimal effect on the FTI and CF for individual drugs in both primary and secondary analyses.
3.3.1 Class analyses
Opioids (excluding heroin) had a significantly greater relative lethal toxicity than the median of all drugs, with a FTI of 40.3 (35.2–45.4) deaths/DDD/100 years and CF of 12.4% (11.0–13.9) overall (Figure 3A). Of the opioid class, CIs for dextropropoxyphene, pethidine and pholcodine were particularly wide and so were omitted from Figure 3. The FTI for methadone appears to be substantially greater than other opioids but this is associated with under capture of dispensings and so difficult to interpret (see limitations). The FTI and CF for methadone (146.7 [107.8–185.7] deaths/DDD/100 years and 29.4% [22.7–36.0]), morphine (85.4 [62.1–108.6] deaths/DDD/100 years and 36.9 [29.0–44.8]) and fentanyl (62.5 [29.5–95.5] deaths/DDD/100 years and 47.1 [29.0–65.3]) were significantly greater than the median of other opioids. Fentanyl appeared to have a relatively higher CF than FTI (Figure 3B). Conversely, oxycodone appeared to have a relatively higher FTI than CF (Figure 3B). Prescribed opioids appear to have similar CF in multiple compared to single agent overdose. However, methadone, oxycodone and morphine may be more toxic in multiple agent deaths as measured by FTI (Figure 4A,B).
Antidepressants had a low relative fatal toxicity, with an overall FTI and CF of 2.5 (2.0–3.0) deaths/DDD/100 years and 2.5% (2.0–3.0) respectively. However, TCAs had a much higher FTI and CF (14.51 (9.74–19.29) deaths/DDD/100 years and CF 7.1% [4.8–9.3] respectively) compared to other subclasses of antidepressants (Figure 3C). Overall FTI and CF for antipsychotics was 7.9 (5.5–10.3) deaths/DDD/100 years and 1.72% (1.2–2.24) respectively. The CIs for FTI and CF for individual antipsychotics were generally wide but point estimates for clozapine (50.10 deaths/DDD/100 years and 15.51%) were higher than other antipsychotics and olanzapine had the lowest point estimate for FTI and CF (1.0 deaths/DDD/100 years and 3.3%; Figure 3D). Quetiapine and periciazine had relatively higher FTIs than CFs (Figure 3D). Quetiapine also appeared to be substantially more toxic in multiple agent overdose compared to single (Figure 4D). Benzodiazepines had a similar overall case fatality to antidepressants but a higher FTI (6.7 [5.6–7.8] deaths/DDD/100 years).
Heroin was the most toxic illicit drug with a CF of 26.4% (19.1–33.7) and had a greater CF when taken as a single agent. Amphetamines had a case fatality that was significantly greater than the average CF of all other drugs combined at 8.5% (5.8–11.2; Figure 5).
4 DISCUSSION
In this 16-year population-level study we have summarised poisoning deaths and overdoses within the Greater Newcastle Hunter Area. We have measured the relative toxicity of all drugs taken in overdose in terms of risk of death at the point of prescribing (fatal toxicity index) and at the point of overdose (case fatality) and focused on classes of drugs commonly prescribed and taken in overdose. With some notable exceptions, FTI and CF are well correlated for all deaths and for deaths involving a single agent some drugs and drug classes appear to be more toxic when taken in combination with other drugs.
The demographics of people overdosing and dying from overdose in terms of age and sex have remained relatively constant over the last 2 decades.16
Of the classes of interest, opioids were the most toxic in terms of FTI and CF. Differences within the opioid class in terms of FTI and CF may partially be accounted for by differences in potencies at the mu opioid receptor17 but this does not explain why morphine was more toxic than oxycodone. This may be a metabolite issue as heroin is rapidly metabolized to morphine, which may lead to a misattribution of causality.18 While the FTI for methadone is difficult to interpret, the CF was high relative to other opioids, which is consistent with previous studies.19-21 The relative toxicity of opioids in our study was slightly different to a whole of population Finnish study, in which methadone had the highest FTI, followed by oxycodone and then tramadol.19 This may reflect national differences in prescribing practices, drug use (including diversion and misuse) or may be a consequence of differing methodology, as the Finnish study did not attempt to weight multiple drug overdoses to account for causality.
The observation that fentanyl has a higher CF relative to FTI indicates that it may be more toxic at the point of overdose rather than prescribing. Since around 2006, there have been increases in fentanyl prescribing and deaths in Australia, some of which may be accounted for by increasing diversion and illicit use.22, 23 The relatively high toxicity of opioids is concerning, that given opioid dispensing in Australia increased 15-fold (500,000–7,500,000) between 1992 and 2012 with proportionate increases in harms measured by deaths and hospitalizations.24
The finding that TCAs have the highest FTI and CF of all of the antidepressants is consistent with previously studies in which other antidepressants have >10-fold lower lethal toxicity.8, 19, 25 Our finding that clozapine was the most toxic antipsychotics is consistent with previous studies11 and its toxicity profile: causing more sedation and hypotension when taken in overdose, particularly in nonadherent or treatment-naive patients.26 A previous study concluded that the FTI of antipsychotics was inversely related to potency at D2 receptors, implying that actions on other receptors may be more important; however, this analysis did not include atypical agents.11 The finding that olanzapine has the lowest relative toxicity of all antipsychotics was consistent with previous European studies.19, 27 The recent escalation in prescribing of atypical antipsychotics in the Australian community, with high rates of off-label use and polypharmacy28-30 in patients who may be at increased risk of overdose is concerning given their higher than average relative toxicity. This is particularly the case for quetiapine, which is frequently found in psychotropic polypharmacy and appears to be a more toxic in multiple agent overdoses.28 The finding that benzodiazepines have a relatively low fatal toxicity but are more toxic in multiple agent deaths is consistent with previous studies.9, 10, 31
It is unsurprising that heroin is the most toxic illicit drug but the finding that amphetamines have higher than average relative toxicity is of concern given the increasing use of methamphetamine in Australia and globally amongst people who already use illicit drugs.32, 33 Our findings are similar to previous studies that estimated relative toxicity using measures of drug availability (number of users as determined by household surveys, number of seizures by law enforcement agencies and estimates of the market size) as the denominator as well as the propensity of drugs to cause deaths as single causative agents.18, 34 Our observation that the point estimate for cannabis CF was greater that amphetamines and MDMA probably relates to coronial misattribution of causality; cannabis may have been found on postmortem toxicology analysis but is unlikely to cause death.18
4.1 Limitations
Previous studies have found that toxicological cause of death on coronial reports on may not always be accurate. This may be the result of misattribution of causality to metabolites or because of postmortem redistribution causing spuriously high concentrations.27, 35 We attempted to account for the metabolite issue in this study. Furthermore, in the setting of multiple agent poisonings, limited attempts were made by coroners to identify causative agents. For example, paracetamol and morphine could be listed as equally associated with a death—an implausible combination given the different time courses of lethality. This study did not systematically evaluate the quality of the coroner's reports with respect to cause of death and each drug was assumed to be equally weighted in causing death. Some drugs commonly prescribed and taken in overdose, such as pregabalin, were notably absent from coronial reports as they are not part of the standard postmortem toxicology assay.
We assumed that prescribing in the Hunter region reflects national prescribing patterns, and this was found to be the case in previous studies.36 The use of DDD for many of these drugs as measures of exposure, in particular opioids and psychotropics, can be inaccurate if the therapeutic dose range is wide.14, 37 Prescribing also goes un-recorded in ASM data for medicines dispensed within the hospital system or as part of the S100 scheme, leading to a spuriously lower denominator and hence an overestimation of FTI for these medicines. This is particularly the case for buprenorphine and methadone, which are both S100 medicines when prescribed for opioid use disorder. Furthermore, FTI cannot be calculated for nonprescription medicines. In these situations, CF is the preferred estimate of relative fatal toxicity where corresponding overdose data is available. FTI and CF represent an average of all people prescribed a drug or taking the drug in overdose and so do not account for individual level predictors of death at the point of prescribing or overdose.
5 CONCLUSION
This study has allowed the estimation of the lethal toxicity of agents commonly taken in overdose by calculating FTI and CF and has explored the correlation and limitations of these measures. Expanding this study to a larger population would increase the power of this technique but is limited by missing information on overdose numbers. While FTI and CF are reasonably well correlated, the FTI estimates the risk of death from the point of dispensing and so is relevant to medicines policy makers and prescribers. CF estimates risk of death at point of overdose and so is relevant to prescribers when assessing overdose risk, emergency physicians and toxicologists when managing overdoses, and public health policy makers in mitigating overdose risk.
ACKNOWLEDGEMENTS
We would like to acknowledge the National Coronial Information System and the Hunter Area Toxicology Service for the use of their data. Geoff Isbister receives funding from an NHMRC Senior Research Fellowship (1061041) and this research was also partially supported by an NHMRC Program Grant (1055176).
COMPETING INTERESTS
There are no competing interests to declare.
CONTRIBUTORS
N.A.B. conceived the project. J.B., C.E.W. and J.R. were involved in data cleaning and analysis. J.B. prepared the manuscript with feedback from other authors. G.K.I. provided HATS data and gave feedback on the manuscript.
