Materials and Methods

The study was designed as a case-control study nested within a prospective cohort study. Dogs anesthetized or sedated at 117 participating private or referral veterinary practices in the United Kingdom were eligible for inclusion. In the cohort portion of the study, participating institutions used a self-administered questionnaire to record information, including outcome at 48 hours (ie, alive, dead, or euthanatized), on all dogs that were anesthetized or sedated during the study period (June 2002 to June 2004). Anesthesia was defined as chemical restraint sufficient to allow endotracheal intubation. Sedation was defined as chemical restraint insufficient to allow endotracheal intubation, had intubation been attempted, and ranged from mild tranquilization to heavy sedation during which the patient was in sternal or lateral recumbency and was unresponsive to minor or moderate stimuli. Anesthetic-related death was defined as death or euthanasia within 48 hours after termination of anesthesia or sedation, except when death was due solely to the surgical procedure or preexisting condition. Thus, all dogs that died or were euthanatized and for which anesthesia or sedation could not be reasonably excluded as a contributory factor were included. An independent review panel consisting of 3 specialist-level veterinary anesthetists and 1 specialist-level veterinary surgeon classified all deaths as anesthetic-related or not. Deaths for which insufficient information was available for the independent review panel to classify were excluded from consideration for inclusion in the case-control portion of the study, though these deaths were included in the cohort study estimates of risks of anesthetic-related death. Further details relating to the cohort study have been published.¹²

In the case-control portion of the study, dogs that experienced an anesthetic-related death (ie, cases) were compared with dogs that did not die within 48 hours after termination of anesthesia or sedation (ie, controls). Details of the patient, procedure, anesthetic management, and personnel involved were recorded for cases and controls on self-administered questionnaires (Appendix). Intended procedure and duration were recorded in addition to actual procedure and duration. For case dogs that died before the procedure was performed, intended duration was recorded as mean duration for control dogs undergoing procedures in the same category.

For each case dog, 4 control dogs were randomly selected and included in the study. Cohort study questionnaires were returned to the investigators on a monthly basis, and data were entered into a relational database^a and exported to a spreadsheet.^b Cumulative monthly frequencies of anesthetic and sedation events were calculated, allowing for identification of individual anesthetic and sedation events by participating institution and, within institution, by event number each month. Controls were identified on the basis of participating institution, procedure number of the day, day of the week, and week of the month on the basis of the previous month's data, and participating institutions were asked to complete questionnaires for controls shortly after the anesthetic or sedation event had taken place.

A pilot study was undertaken between June and October 2002 to refine a priori hypotheses, test methods and tools for data collection, and check sample size calculations. Sample size calculations indicated that with a case-to-control ratio of 1:4, approximately 140 to 150 cases would be required to detect risk factors with a prevalence of 5% to 15% in control dogs and ORs of 2.0 to 2.5 with a power of 80% and ≤ value of 0.05.

Statistical analysis—Univariable analyses (ie, χ² or Fisher exact test for categorical data and Student t test or Mann-Whitney U test for continuous data^13,14) were used to test for associations between potential risk factors and anesthetic-related death. Factors for which the P value in these univariable analyses was < 0.2 and factors that were considered to be biologically important were then analyzed by means of multivariable, mixed-effects logistic regression with a manual forward-selection approach.¹⁴ Continuous variables were evaluated for linearity and higher order associations and were examined as fractional polynomials.¹⁵ Institution identity was treated as a random effect in the mixed model to account for clustering of outcome. First-order interactions were assessed, and final model fit was examined with the Hosmer-Lemeshow test statistic and the delta B and delta deviance influence diagnostic statistics.¹⁴

Characteristics of a sample of 20% of nonresponding controls were compared with characteristics of responding controls to assess likely representativeness of the controls. All analyses were performed with standard software.^c Values of P ≤ 0.05 were considered significant.

Results

Information was obtained on 98,036 anesthesia and sedation events between June 2002 and June 2004. A total of 148 anesthetic-related deaths (cases) occurred during this period. Additional details regarding timing and cause of deaths have been reported.¹²

For analysis of risk factors, information was obtained for 148 cases and 487 randomly selected controls. Median age of case dogs was 8.0 years (interquartile [25th to 75th percentile] range, 5.0 to 11.5 years); median age of control dogs was 5.6 years (interquartile range, 1.6 to 9.0 years). Forty-seven percent (69/148) of case dogs and 49% (238/487) of control dogs were male. In univariable analyses, dogs in the toy, utility (miscellaneous breeds of dogs mainly of a nonsporting origin, including the Bulldog, Dalmatian, Japanese Akita, and Poodle), and pastoral (herding dogs associated with working cattle, sheep, reindeer, and other cloven-footed animals) groups were overrepresented in the case group, compared with the control group (Table 1). Individual breeds that were overrepresented included Yorkshire Terrier, West Highland White Terrier, Jack Russell Terrier, German Shepherd Dog, and Cocker Spaniel. Neutering was the most common procedure performed, followed by minor soft tissue surgery, diagnostic procedures, major surgical procedures, and dental surgery (Table 2). Twenty-eight case dogs died before a procedure could be undertaken and 2 control dogs did not undergo a procedure. Median actual procedure duration was 60 minutes (interquartile range, 30 to 100 minutes) for the case dogs and 45 minutes (interquartile range, 30 to 65 minutes) for the control dogs.

For 8 of the 148 (5%) case dogs and 36 of the 487 (7%) control dogs, sedation was used. Most often, sedation was achieved with a combination including medetomidine (3/8 [38%] case dogs and 29/36 [81%] control dogs). In the remaining 140 (95%) case dogs and 451 (93%) control dogs, anesthesia was induced. Most dogs were premedicated prior to anesthesia (103/140 [74%] case dogs and 423/451 [94%] control dogs), principally with combinations incorporating acepromazine (64/140 [46%] case dogs and 376/451 [83%] control dogs), medetomidine (4/140 [3%] case dogs and 26/451 [6%] control dogs), or a benzodiazepine or opioid (35/140 [25%] case dogs and 21/451 [5%] control dogs). Anesthesia was induced most often with propofol (112/140 [80%] case dogs and 319/451 [71%] control dogs) or thiopentone (14/140 [10%] case dogs and 120/451 [27%] control dogs). Mask induction was used in 12 of 140 (9%) case dogs and 6 of 451 (1%) control dogs, and other combinations were used for anesthetic induction in 2 of 140 (1%) case dogs and 6 of 451 (1%) control dogs. In the anesthetized dogs, an endotracheal tube was placed in 137 of 140 (98%) case dogs and 447 of 451 (99%) control dogs. Anesthesia was maintained with isoflurane (104/140 [74%] case dogs and 396/451 [88%] control dogs), halothane (19/140 [13%] case dogs and 41/451 [9%] control dogs), sevoflurane (5/140 [4%] case dogs and 10/451 [2%] control dogs), and injectable anesthetics (12/140 [9%] case dogs and 4/451 [1%] control dogs).

One hundred sixteen of the 148 (78%) case dogs and 462 of the 487 (95%) control dogs were breathing spontaneously during the anesthetic or sedation procedure. Oxygen supplementation was provided for 141 of 148 (95%) case dogs and 453 of 487 (93%) control dogs. Intravenous fluid therapy was administered to 108 of 148 (73%) case dogs and 93 of 487 (19%) control dogs. One hundred ten of 148 (74%) case dogs and 416 of 487 (85%) control dogs were monitored by a qualified veterinary nurse (qualified veterinary nurses undergo a 2-year period of vocational training, which is assessed at work and through examination, and a Certificate in Veterinary Nursing is awarded by the UK Royal College of Veterinary Surgeons), 2 of 148 (1%) case dogs and 24 of 487 (5%) control dogs were monitored by an unqualified veterinary nurse, 33 of 148 (22%) case dogs and 40 of 487 (8%) control dogs were monitored by a veterinarian other than the person undertaking the procedure, and 3 of 148 (2%) case dogs and 7 of 487 (2%) control dogs were monitored by the same person performing the procedure. Pulse oximetry was used intraoperatively in 80 of 148 (54%) case dogs and 246 of 487 (51%) control dogs, electrocardiography was used in 24 of 148 (16%) case dogs and 55 of 487 (11%) control dogs, capnography was used in 22 of 148 (15%) case dogs and 49 of 487 (10%) control dogs, and blood pressure monitoring was used in 16 of 148 (11%) case dogs and 47 of 487 (10%) control dogs. Patient temperature was measured postoperatively in only 17 of 148 (11%) case dogs and 68 of 487 (14%) control dogs.

In the multivariable model, ASA physical status, urgency of the procedure, type of procedure (ie, major vs minor), age, body weight, the combination of agents used for anesthetic induction and maintenance, and intended duration of the procedure were significantly associated with the odds of anesthetic-related death (Table 3). An increase in ASA physical status from ASA 1 or 2 to ASA 3 or from ASA 3 to ASA 4 or 5 was associated with a 6.6-fold increase in the odds of anesthetic-related death, and an increase in urgency of the procedure from scheduled to urgent or from urgent to emergency was associated with nearly a 3-fold increase in the odds of anesthetic-related death. Dogs requiring major procedures were approximately 5 times as likely to experience an anesthetic-related death as were dogs requiring minor procedures. Dogs that were ≥ 12 years old were approximately 7 times as likely to experience an anesthetic-related death as were young adult dogs (0.5 to 8 years). Dogs weighing < 5 kg (11 lb) were nearly 8 times as likely to experience an anesthetic-related death as were dogs that weighed ≥ 5 to < 15 kg (11 to 33 lb). Induction of anesthesia with injectable agents and maintenance with halothane was associated with a 5.8-fold increase in the odds of anesthetic-related death and induction and maintenance of anesthesia with inhalant anesthetics was associated with a 5.9fold increase in the odds of anesthetic-related death, compared with induction of anesthesia with injectable agents and maintenance with isoflurane. The odds of anesthetic-related death following use of injectable agents for sedation were similar to the odds of death following induction of anesthesia with injectable agents and maintenance with isoflurane. However, the odds of anesthetic-related death following use of injectable agents for anesthesia were substantially increased, compared with the odds following induction of anesthesia with injectable agents and maintenance with isoflurane, primarily because this included dogs that died during anesthetic induction prior to maintenance of anesthesia (8/12 dogs that only received injectable agents for anesthesia died during anesthetic induction). There was an insufficient number of dogs given sevoflurane to make meaningful comparisons. An increase in the intended duration of the procedure by 1 minute was associated with a 1.006-fold increase in the odds of anesthetic-related death. Although the P value associated with intended duration of the procedure was 0.07, this factor was retained in the model on biological grounds and because its inclusion improved the fit of the model, as assessed with the delta E and delta deviance diagnostic statistics. After adjustment for body weight, breed was not retained in the final multivariable model.

Response rates for the case and control groups were good. There were 287 deaths recorded during the study period, and information was sufficient for all but 15 for the independent review panel to classify them as anesthetic-related or not. Similarly, information was requested for 616 control dogs, and questionnaires were returned for 503, of which 16 were excluded. Nonresponding control dogs were comparable to responding control dogs in regard to ASA physical status, intended procedure, and age (Table 4), although nonresponding control dogs were more likely to have required urgent or emergency procedures than reported control dogs. Cases were clustered at the clinic level (P = 0.06), and although the P value was higher than our cutoff, this factor was retained as a random effect in the multivariable model on biological grounds. The fit of the multivariable logistic regression model was good as assessed by the Hosmer-Lemeshow goodness-of-fit statistic (P = 0.80), the delta E diagnostic statistic (all delta B < 1.0), and the delta deviance diagnostic statistic (only 5 covariate patterns > 6.0).

Discussion

Results of the present study suggested that patient health status (ie, ASA physical status); age; weight; and urgency, complexity, and anticipated duration of the procedure were significantly associated with the odds of anesthetic-related death in dogs undergoing anesthesia or sedation. Preoperative assessment of these factors should allow veterinarians to identify those dogs at greatest risk of anesthetic-related death. Results also suggested that care should be exercised with mask induction of anesthesia and that isoflurane would appear preferable to halothane for maintenance of anesthesia. Further work in the form of randomized controlled trials is needed to verify these observations.

To our knowledge, the present study represents the first large-scale, prospective, multicenter study of anesthetic-related deaths in dogs in the past 10 to 20 years. Although the study was restricted to the United Kingdom, risk factors that were identified are probably relevant internationally. Findings should encourage better patient preparation, more extensive anesthetic monitoring, and more intensive perioperative management, particularly in dogs at high risk, to decrease the risk of anesthetic-related death. Highlighting major factors associated with increased odds of this complication could allow a reevaluation of best practices to further reduce the risk of death.

Patient health status, as described by ASA physical status, was particularly important as a risk factor for anesthetic-related death in the present study. Although the ASA physical status categories were created for use in human medicine,¹¹ our results provide further support for the relevance of these categories in other species when assessing the risks of anesthesia. An association between health status and anesthetic-related death in animals has been documented previously^{1,3,8–10,12,16,17} and is consistent with published findings for people.^7,18–24 Preexisting medical conditions that increase ASA physical status may reduce the therapeutic index of administered anesthetics, predispose to cardiopulmonary depression, and depress other physiologic functions.

Procedural urgency was also found to be a risk factor for anesthetic-related death in the present study. Risk of anesthetic complications has previously been associated with increasing urgency in humans, horses, and cats,^{7,16,19,21,24–27} and this likely reflects the ability to assess and stabilize patients preoperatively and, because of the tendency for urgent procedures to be performed outside of normal working hours, staffing levels and personnel fatigue. Greater attention to preoperative assessment of patient health status and procedural urgency and improved stabilization prior to induction of anesthesia could potentially reduce the risk of anesthetic-related death.

Interestingly, the increase in odds of anesthetic-related death was apparent only in the oldest dogs in the present study, such that dogs ≥ 12 years old were nearly 8 times as likely to die as were young adult dogs, whereas the odds of death among dogs 8 to 12 years old were similar to the odds among younger dogs. Qualitatively, these findings agree with findings in a previous study⁹ involving dogs examined at a referral practice and in studies involving horses^17,27 and humans.^{7,19,20,22,23} Old patients may be more susceptible to the depressant effects of anesthetics, to hypothermia because of impaired thermoregulatory mechanisms, and to prolonged recovery times because of reduced metabolic function and hypothermia.^28–30 Thus, particular care should be taken when anesthetizing these patients.

The increase in the odds of anesthetic-related death among small patients in the present study was consistent with findings in cats¹⁶ and children.³¹ Smaller patients could be more prone to drug overdosage, hypothermia, and perioperative difficulties (eg, difficulties with IV catheter placement or endotracheal intubation). This finding was unlikely to reflect major confounding by age, as the association with weight was adjusted for age in the multivariable model. Interestingly, breed was associated with anesthetic-related death in univariable analyses, but after adjustment for weight, this association was not retained in the multivariable model, suggesting a major aspect of the risk of anesthetic-related death in particular breeds could be related to their small size. However, small size alone cannot account for the association between anesthetic-related death and utility or pastoral breeds, and caution with anesthesia of these breeds, particularly brachycephalic breeds, may be advisable. This is consistent with results of previous studies^1,3 in which terriers and brachycephalic breeds, such as Pekinese and Bulldog, were highlighted as having a high risk of complications.

The higher odds of anesthetic-related death among patients requiring major versus minor procedures were also consistent with results of studies involving cats,¹⁶ horses,^17,27 and people.^19,24,25 More complex and invasive procedures were likely to impose greater stresses on patient physiology. Thus, when assessing patient risk prior to anesthesia, the planned procedure's complexity should be considered. Additionally, intended duration of the procedure was retained as a risk factor in the multivariable model, suggesting that independent of the type of procedure, an increase in procedure duration was associated with an increased risk of anesthetic-related death. Anesthetic duration has been reported as a risk factor for death in humans^24,32,33 and horses²⁷ undergoing anesthesia. Longer procedures would expose patients to potentially longer periods of physiologic compromise, greater risks of hypothermia, and larger fluid losses.³⁴

The combination of agents used for anesthetic induction and maintenance was the only drug-related factor retained in the final multivariable model in the present study, with the major finding being approximately 6-fold increases in the odds of anesthetic-related death following induction with injectable agents and maintenance with halothane and following induction and maintenance with inhalant anesthetics, compared with induction with injectable agents and maintenance with isoflurane. The substantial increase in odds of death with the use of injectable agents alone for anesthesia likely reflected inclusion of patients that died before an inhalant agent could be delivered. Of 12 dogs that were anesthetized with injectable agents alone, 8 died during induction of anesthesia and an additional 3 died before a procedure was performed. By comparison, only 4 control dogs were anesthetized with injectable agents alone. Hence, a true association with use of injectable agents was difficult to assess.

The increase in the odds of anesthetic-related death for dogs in which anesthesia was induced with injectable agents and maintained with halothane, compared with the odds in dogs in which anesthesia was induced with injectable agents and maintained with isoflurane, contrasts with results of a previous study³ in which halothane was associated with reduced odds of complications in small animal practice but was consistent with results of studies^17,35 in which isoflurane was associated with reduced odds of complications in young horses. The reduced odds of complications associated with halothane in the previous study³ may have resulted from residual confounding by health status (ie, veterinary surgeons were less likely to choose halothane for sicker patients). In addition, although results were adjusted for ASA grade, only a binary variable was used (ASA grade 1 or 2 vs ASA grade 3, 4, or 5) and the association was with nonfatal complications only. Although isoflurane induces greater respiratory depression and vasodilation than halothane, it causes less direct myocardial depression and sensitizes the heart less to catecholamine-induced arrhythmias. Thus, on balance, isoflurane would appear to cause less overall cardiovascular depression.^36–45

The increase in the odds of anesthetic-related death following induction and maintenance of anesthesia with inhalant anesthetics, compared with the odds following induction with injectable agents and maintenance with isoflurane, was consistent with previous studies^1,27 involving small animals and horses in which mask induction tended to increase the risk of death. Induction of anesthesia with inhalant anesthetics can be stressful, necessitates a longer period prior to tracheal intubation, and causes greater cardiopulmonary depression if excessive depth is reached prior to endotracheal intubation.³⁴ In recent work in dogs, mask induction was associated with greater cardiopulmonary depression^d and was less smooth⁴⁶ than use of injectable combinations. Interestingly, sedation appeared to be associated with similar odds of anesthetic-related death, compared with induction with injectable agents and maintenance with isoflurane, in the present study, suggesting that sedation may be an appropriate alternative to general anesthesia in some settings.

Importantly, the major injectable agent for anesthetic induction in the present study was not associated with the risk of anesthetic-related death. There has been a perception in veterinary practice that propofol is a safe drug,^47,48 but the cardiopulmonary effects of propofol and thiopentone are similar.^34,49–55 We also did not identify a significant association between the premedication administered and anesthetic-related death. The odds of death following premedication with medetomidine were similar or lower than the odds following premedication with acepromazine, when the association of premedication with anesthetic-related death was adjusted for ASA physical status and other variables in the final model (adjusted OR, 0.4; 95% CI, 0.1 to 2.1), whereas giving no premedication or giving only an opioid and benzodiazepine combination tended to be associated with increased odds of death. However, this association was not significant (P = 0.26) and was not retained in the final model. Acepromazine premedication has been associated with reduced risk of complications in small animals and horses, compared with no premedication.^1,3,27,56 The fact that we did not identify a difference in the odds of anesthetic-related death following premedication with medetomidine versus acepromazine was interesting, particularly because an older D₂-adrenoceptor agonist, xylazine, has been associated with an increased risk of death in a number of studies^1,3,57 involving small animals. The cardiopulmonary effects of the D₂-adrenoceptor agonists are well known and include transient hypertension followed by hypotension, bradycardia, increased systemic vascular resistance, reduced cardiac output, and minimal respiratory depression.^58–66 Additionally, xylazine has been found to sensitize the heart to catecholamine-induced arrhythmias during halothane anesthesia,^67,68 although medetomidine has not.⁶⁹ These cardiovascular differences, combined with a greater awareness of the physiologic effects of D₂-adrenoceptor agonists and how to use them, may be the basis for the lack of increased risk with medetomidine, compared with acepromazine. However, additional work is needed to evaluate this further before major conclusions are drawn on the relative safety of individual premedications.

The validity of the present study rested on consistent application of a clear definition of anesthetic-related death and use of an independent review panel to assess all potential cases. The high response rate for potential cases indicated that cases included in the study were representative of anesthetic-related deaths occurring in the study population. Controls were randomly selected from the cohort of anesthetized and sedated dogs to reduce the potential for selection bias. The response rate for controls was good, and comparisons between responders and nonresponders suggested the controls were largely representative of the population under study. That nonresponding controls were more likely to have required urgent or emergency procedures (42.8%) than responding controls (14.2%) may have been due to the small sample size of nonresponding controls evaluated, but even if this was due to selection bias, it was unlikely to have substantially affected our results, as the odds of death would still have been much higher with urgent or emergency procedures, compared with scheduled procedures. Misclassification of exposure history was minimized by checking data against the anesthetic record when available and assessing the plausibility of data. Recall bias was minimized by requesting that questionnaires for controls be completed soon after the procedures were undertaken, such that both case and control questionnaires were completed in the immediate period after the anesthetic or sedation event and control selection could not influence exposure status.¹³