Learning from critical care management of sheep receiving extracorporeal membrane oxygenation 1 for smoke-induced acute lung injury as a tool for processing large clinical datasets 2 3

Background: Many successful therapies developed for human medicine involve animal experimentation. With competition for public research funding and career advancement opportunities, animal studies focused on the translational potential may not sufficiently document unexpected outcomes. Such studies often have hastily developed methods with ad hoc modifications including the use of additional animals, leading to considerable amounts of idle, unprocessed data that could be used to advance veterinary science, or to refine the base animal model. Sheep, for example, are poorly understood models of intensive care and therefore, any experimental data arising from them should be interpreted with care. The hypothesis was that there is little information describing the development of methods of physiological data processing in multifaceted sheep models of intensive care and the author aimed to develop a suitable data processing method and to analyse the data, once processed. Methods: Data from 19 adult mechanically ventilated ewes undergoing intensive care in a previous study evaluating a form of extracorporeal life support (treatment) for acute lung injury were used to develop a comprehensive method for processing manual and electronically gathered clinical observations. Eight sheep were injured by acute smoke inhalation prior to treatment (injured/treated), while another eight were not injured but treated (uninjured/treated). Two sheep were injured but not treated (injured/untreated), while one received room air instead of smoke as the injury, and was not treated (placebo/untreated). The data were then analysed for 11 physiological categories and compared between the two treated groups. Results: The analysis revealed that compared with the baseline, treatment contributed to and exacerbated the deterioration of pulmonary pathology by reducing lung compliance and PaO2/FiO2 ratio. The oxygen extraction index changes mirrored those of the PaO2/FiO2 ratio. Decreasing coronary perfusion pressure predicted the severity of cardiopulmonary injury. Conclusions: These novel observations could help in understanding similar pathology such as that which occurs in smoke inhalation animal victims of house and bush fires. A similar data processing method could be used when evaluating the effectiveness of other clinical interventions such as potentially reversible aspiration pneumonia secondary to tick paralysis in veterinary patients. Keywords: Sheep, critical care, smoke-induced acute smoke inhalation injury, extra-corporeal life support, lung compliance, PaO2/FiO2 ratio


48
During multifaceted experiments involving intensive care in large animal models in translational research, 49 information regarding animal monitoring is often collected with varying accuracy, scope, and end-user 50 applications. Data collection can be manual, electronic, or both [1-3]. Manually input data can include 51 subjectively scored end-points such as the plane of anaesthesia, and objective data such as heart rate or 52 breaths per minute. Depending on the goals of the study, some information may be used to validate or test 53 novel therapies, or to understand and refine existing treatments. In some cases, experimental information 54 may be gathered for scientific curiosity or for "classified" use, and outcomes may never be publicly available,

73
The overall goal was to provide useful information relevant to the sheep model, itself, and to those 74 interested in large animal experimentation and veterinary medicine, generally. The specific objectives were: 75 1) use the raw data from the sheep model study to create a data management system for tabulating large 76 data sets from human studies using animal models and, 2) analyse that data to provide biological information 77 that is not currently available for sheep receiving ECLS following smoke-induced acute lung injury.

78
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

79
The study was carried out at the purpose-built Medical Engineering Facility of Queensland University of

99
The source study involved 64 sheep, comprising eight experimental groups of eight sheep based on 100 the study's multiple objectives, subsequent modifications, and later addition of experimental controls. The 101 experimental groups were classified based on: the duration of treatment (2 and 24 hours; E2H and E24H); 102 treatment after smoke inhalation (injury) for 2 and 24 hours (SE2H and SE24H); and treatment after smoke 103 inhalation and transfusion with fresh or stored (aged) blood (SEF24H or SEA24H), respectively. Two 104 additional groups included one group receiving smoke inhalation injury but no treatment (SC24H), and 105 another group that inhaled room air only as the injury (placebo) and no treatment (C24H). Data from sheep 106 involved in the treatment and transfusion studies were not included in the analysis in this study because 107 these data were beyond the scope of this study. Nineteen sheep were included in the present study; data 108 were analysed for 16 sheep with robust data (E24H and SE24H), and included, but not analysed, for three 109 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; from groups SC24H and C24H (early observational data). A systematic approach was developed for 110 processing the data.

112
Manually acquired physiological data workflow 113 A clone of the master manual data entry spreadsheet was created by removing the formatting and formulas.

114
Several members of the sheep ECMO research team inspected data repeatedly for errors to ensure that all 115 columns, rows, time points, and data points had been copied correctly, including number formats (Figure 1).

116
Redundant columns were removed and data were aligned to experimental time points (Figure 2). While 117 maintaining the same experimental time point header, data from the table in Figure 2 were split and grouped 118 into the following categories: ventilator settings, blood pressure and haemodynamics, fluids and urine output, 119 arterial blood gas values, activated clotting time, anaesthetics, anticoagulants, and ECLS circuit 120 observations.

122
Electronically acquired physiological data workflow 123 Raw electronically acquired physiologic monitoring data were inspected for completeness. The data 124 comprised 36 time points: ECLS pump time (min); time of day (h); electrocardiograph (heart rate); arterial 125 blood pressure (mean, systolic, diastolic, heart rate); central venous pressure (mean); pulmonary artery 126 pressure (mean, systolic, diastolic); oxygenator pressure (pre-and post-); capnography (end-tidal carbon 127 dioxide (etCO 2 ), respiratory rate); pulse oximetry (SpO 2 , heart rate); ECLS pump (flow rate, speed); ventilator 128 (mode, frequency, oxygen, pressure control, inspiratory volume, expiratory volume, expiratory minute 129 volume, pressure maximum, mean pressure, positive end-expiratory pressure, plateau pressure, inspiratory 130 resistance, expiratory pressure, pulmonary compliance, inspiratory flow); mixed venous oxygen saturation 131 (SvO 2 ); and continuous cardiac output (CCO) (Figure 3). The yellow line in Figure 3 indicates the baseline 132 time point and the grey line represents the smoke inhalation time point. It is important to note that there may 133 or may not have been any data at any given point in time.

134
The electronically acquired physiological monitoring data were inspected for errors and cleaned to provide 135 data for downstream analysis (Figure 4).

137
Pre-data analysis checks 138 Data were then subjected to further integrity checks. An important step was to make a plot of data versus 139 time together with descriptive statistics for all data points in the grouped data. At the time of data processing,

140
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; the "Descriptive Statistics" tool of Microsoft® Excel 2010 (Microsoft Corporation) did not complete analysis 141 with missing values. Therefore, the data to be analysed were selected and then the "GoTo" tool (F5) was 142 used, and "Special", "Blanks", and thereafter, "OK" were selected to identify blanks. The blanks were deleted 143 by positioning the cursor in the blank cell and using the space bar to clear the cell (the "Delete" or 144 "Backspace" keys did not remove the blanks).

145
After artefact removal and integrity checks, data for individual sheep were placed into six categories: 146 activated clotting time; anaesthetics + inotropes and anticoagulants; arterial blood gas values; blood 147 pressure + ventilation and haemodynamic data; calculated respiratory + haemodynamic variables; and fluids 148 and urine production. Using specially written macros, data were extracted from each experiment and 149 grouped by parameter corresponding to experimental time points. All sheep treatment data were filed by

160
To meet the second objective, data from the groups, uninjured/treated and injured/treated groups were 161 analysed. The means, medians and standard deviations of the weights of the sheep, where applicable, were 162 tabulated. The physiological parameters of the groups were charted and compared against each other using 163 one-way analysis of variance (ANOVA), where appropriate. Parameters between groups were compared 164 using a paired two-tailed t-test. All p-values were two-sided and p < 0.05 was considered statistically 165 significant. All statistical calculations were performed using GraphPad PRISM 6 software (GraphPad

169
The biodata of the sheep that were used in the current analysis are presented in Table 1. The weights of the 170 uninjured/treated sheep, unlike the injured/treated group, did not pass the D'Agostino-Pearson omnibus 171 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; normality test; however, there was no significant difference in the weights of the sheep between groups.

192
The pCO 2 in all but the uninjured/treated sheep increased initially before plummeting sharply, 193 forming a shallow trough corresponding to 1 hour after the start of treatment, followed by a slight increase 194 before stabilising in all sheep.

195
There was a gradual decrease in pO 2 in the treated groups of sheep from baseline before

202
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; 8 The fraction of oxyhaemoglobin (FO 2 Hb) decreased sharply with the lowest reading at 5 minutes 203 post-injury before returning to near baseline levels within an hour of starting treatment. The injured/treated 204 sheep had a considerably deeper trough in FO 2 Hb level and there was a significant difference (p = 0.046) 205 between troughs. There was no change in FO 2 Hb for the uninjured sheep.

206
The fraction of carboxyhaemoglobin (FCOHb) increased sharply from baseline, peaking at 207 approximately 5 minutes post-injury and decreased sharply thereafter to the start of treatment before 208 gradually returning to near baseline levels at approximately 6 hours post-start of treatment in the injured 209 sheep. The injured/treated sheep had a higher peak FCOHb than the injured/untreated sheep, although the 210 difference was not significant. There was no change in FCOHb for the uninjured sheep.

219
The blood sodium concentration [Na + ] was relatively stable and there were no significant differences 220 between groups.

221
There was an initial decrease in the blood calcium [

232
Overall, the anion gap decreased gradually, reaching a relatively gentle slope at approximately 6 233 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; hours from the start of treatment and did not change significantly, thereafter. There was a gradual decrease 234 in anion gap from baseline during the course of the experiments and there was no significant difference in 235 anion gap between the uninjured/treated and injured/treated groups.

238
Although there was an increase in blood glucose level [Glu] for the injured/treated sheep after 6 hours of 239 treatment, the change was not significant.

240
There was an initial decrease in lactate levels [Lac] 6 hours after the start of treatment, followed by a 241 gradual increase for the injured sheep, especially for the injured/treated group. There was no significant

262
The mean pulmonary artery pressure (MPAP) increased gradually, with the injured/treated group 263 having a consistently higher MPAP. There was no significant difference in MPAP between the 264 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

265
There was an initial, subtle increase in central venous pressure (CVP) that peaked at approximately 266 1 hour post-injury followed by a decrease that stabilised at approximately 1 hour post-start of treatment. CVP 267 levels in the injured/treated and placebo/untreated sheep were consistently higher and lower, respectively,    After an initial increase in systemic vascular resistance index (SVRI) to approximately 1 hour after 291 the start of treatment, SVRI began to decrease in all experimental groups before plateauing after 12 hours of 292 treatment followed by a gentle increasing trend until the end of the experiments. SVRI in the injured/treated 293 group was consistently below that of the other groups during treatment while that of the injured/untreated 294 group was correspondingly higher. There was no significant difference in SVRI between the groups.

295
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; Pulmonary vascular resistance index (PVRI) remained close to baseline levels for all of the groups 296 until 1 hour after the start of treatment when that of the injured groups progressively increased while that of 297 the uninjured groups remained lower with a subtle decrease to 6 hours post-treatment. PVRI in the 298 placebo/untreated sheep remained near baseline levels and the lowest throughout the course of the 299 experiment.

300
After a small peak corresponding to the start of treatment, right ventricular stroke work index 301 (RVSWI) in the uninjured sheep gradually increased while that of the injured sheep decreased. There was a 302 significant difference (p = 0.0196) in RVSWI gap between the uninjured/treated and injured/treated groups.

303
RVSWI in the placebo/untreated group remained high while that of the injured/treated group was consistently 304 the lowest.

305
Left ventricular stroke work index (LVSWI) gradually increased in the uninjured/treated and 306 placebo/untreated groups, before plateauing after 12 and 18 hours of treatment, respectively, while LVSWI in 307 the injured/untreated and injured/treated groups of sheep decreased before plateauing at 12 hours and 308 trending upward after 18 hours of treatment. LVSWI in the placebo/untreated group remained consistently 309 higher than in the other groups while that of the injured/treated group was consistently the lowest.

315
There was an initial subtle decrease in arterial oxygen content (C a O 2 ) from baseline in all groups 316 before a sustained increase in the injured/untreated group, a steady level in the placebo/untreated sheep, 317 and a sharp trough in the injured/treated and uninjured/treated groups. Following the trough, the C a O 2 of the 318 injured/treated group gradually returned to baseline levels while that of the uninjured/treated group continued 319 on a downward trend. There was a significant difference (p < 0.0085) in C a O 2 between the uninjured/treated 320 and injured/treated groups.

321
There was a slight decrease in the oxygen delivery index (DO 2 I) in all groups to 1 hour of treatment 322 before a further marked decrease, except for the placebo/untreated sheep. There was a significant 323 difference (p = 0.0013) in DO 2 I between the uninjured/treated and injured/treated groups. The injured/treated 324 group had the lowest DO 2 I compared with the other groups while the placebo/untreated sheep maintained 325 the highest DO 2 I profile.

326
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; The oxygen extraction index (O 2 EI) decreased in all groups before plateauing at approximately 6 327 hours after the start of treatment. There was a significant difference (p = 0.0247) in O 2 EI between the 328 injured/treated and uninjured/treated groups. O 2 EI in the injured/treated and injured/untreated groups was 329 consistently lower and higher, respectively, compared with those of the other groups.

336
The injured/untreated and injured/treated groups produced the least and most urine on average, 337 respectively. There was no significant difference in urine output between the uninjured/treated and 338 injured/treated groups.

341
There was a significant difference (p < 0.0001) in the amount of alfaxalone required between the 342 uninjured/treated and injured/treated groups. The uninjured/treated group required more alfaxalone on 343 average and the injured/untreated group required the least amount on average. Ketamine requirements 344 differed between groups, with the injured/untreated group requiring the highest amount on average and the 345 injured/treated group requiring the least. There was no significant difference in the quantities of ketamine 346 required between the uninjured/treated and injured/treated groups but significant differences in midazolam 347 requirements occurred between the uninjured/treated and injured/treated groups (p = 0.0067).

350
There were no significant differences in heparin infusion doses between the uninjured/treated and 351 injured/treated sheep. Heparin requirements for the placebo/untreated group were the lowest. Activated CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; 13 There were significant differences in the ECLS pump speed, blood flow, and pressure differential between 358 the uninjured/treated and injured/treated groups. Pump speed, blood flow, and pressure differential were 359 significantly different (p = 0.0022), (p = 0.0095) and (p = 0.0041), respectively, between the two groups 360 receiving ECLS. These parameters in the uninjured/treated group were consistently higher than those of the 361 injured/treated group.

374
The results of this study agree with and confirm earlier preliminary observations that ECLS causes a 375 decrease in pulmonary compliance over time [9]. It was expected that the injured sheep would have 376 relatively lower SpO 2 readings compared with the other groups because of episodes of hypotension with 377 hypoxemia, which can affect pulse oximeter function [14]. The relatively low etCO 2 in the injured sheep 378 suggested that the sheep may have hyperventilated, the causes of which were evaluated with respect to 379 reactive oxygen species or superoxide dismutase activity by a team from the source study [15,16].

380
The relatively low blood pH in the injured/treated sheep suggested that the sheep tended to 381 metabolic acidosis as the same group of animals also had low etCO 2 . This also means that there was no 382 respiratory component contributing to the observed acidosis. The low pCO 2 in the uninjured/treated sheep 383 could be a result of hyperventilation and the high pCO 2 in the injured/untreated sheep suggested that CO 2 384 clearance was curtailed by injury.

385
The treatment of the sheep contributed to lung injury by causing deterioration of pO 2 . The low pO 2 386 translated to low partial arterial oxygen pressure/inspired oxygen fraction (PaO 2 /FiO 2 ) ratio, which was much 387 worse in the injured sheep. This finding showed that ECLS contributed to the deterioration of the PaO 2 /FiO 2 388 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; ratio in the injured/treated group of sheep, a novel finding that was also unexpected in the primary study.

389
The relatively higher levels of [Hb]   concern with sub-optimal rumen function leading to loss of its buffer effect and increasing numbers of

417
The elevated [Base (ecf)] above +2 mmol/L for most of the first 12 hours in the placebo/untreated 418 and injured/untreated sheep suggested that the sheep were metabolically alkalotic [29] before returning to 419 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; normal levels. The relatively low [Base (ecf)] (less than −2 mmol/L) was consistent with HCO 3 loss and the 420 tendency to metabolic acidosis [29] in the injured/treated sheep. The marked decrease in [HCO 3 -] in the 421 injured sheep was consistent with metabolic acidosis and was more severe in the injured/treated group,

422
suggesting that ECLS was a contributing factor.

423
The resting HR of sheep is 50-80 beats/min [22]. In a study that instrumented conscious sheep, the 424 baseline heart rate was registered as 106 ± 9 beats/min

436
There was a benefit of ECLS treatment for SvO 2 as it remained high for both the injured/treated and 437 uninjured treated groups. The consistently low SvO 2 in the injured/untreated group was expected because of 438 the slightly reduced cardiac output in this group; however, this level of SvO 2 was still higher than that

447
The reduction in coronary perfusion pressure in the injured/treated, and to a certain extent the 448 uninjured/ treated sheep, suggested that ECLS contributed to the decrease in CPP, in addition to smoke 449 injury. CPP is an indicator of myocardial perfusion and has been proposed as a drug target during 450 . CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; resuscitation [36]. The observations in the present study support the suggestion that CPP could be used to 451 predict the severity of injury in sheep.

452
The apparent increase in CaO 2 in the injured sheep could have been due to the relative increase in

453
[Hb] secondary to dehydration. The low DO 2 I in the injured/treated and uninjured/treated groups suggested 454 that ECLS had a contribution, in addition to smoke, based on the relatively higher DO 2 I in the 455 injured/untreated sheep. Interestingly, the O 2 EI had a comparable profile to that of the PaO 2 /FiO 2 ratio, and 456 could also be used to predict the contribution of ECLS to smoke-related injury.

457
The smoke-injured sheep required considerable amounts of intravenous fluids to compensate for the

481
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint . http://dx.doi.org/10.1101/058511 doi: bioRxiv preprint first posted online Jun. 12, 2016; especially for the manually input data. There was also no information about pre-anaesthetic blood tests.

482
An additional limitation relates to the first objective of creating a data management system for tabulating 483 large data sets from human studies using animal models. Because the method has not been validated, it is 484 considered preliminary and further validation studies are required. Also, the numbers of sheep were low and 485 this was especially so in the injured/untreated and placebo/untreated groups, preventing comparisons 486 between the treated and untreated sheep. A further limitation is that cytokine levels, as predictors of lung 487 injury, were not quantified. Using ELISA assays to quantify cytokine levels proved difficult and the cost was 488 prohibitive in the present study. It is partly for this reason that pioneering studies for the development of 489 proteogenomic assays were proposed [44] as an alternative to ELISA to learn from circulating markers of 490 acute inflammation in injured sheep used as models of intensive care, to understand critical illness.

493
The results of this study demonstrated that this preliminary method of raw data processing was effective and

512
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

543
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

561
The author (SC) was solely responsible for the study design, writing the manuscript, analysing and 562 interpreting the data, final approval of the manuscript, and is fully accountable for the work.

573
Jane Charbonneau of Edanz Group, Ltd. for editorial support.

574
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

716
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.

723
. CC-BY-NC-ND 4.0 International license It is made available under a (which was not peer-reviewed) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.