兽研1. 引言
体表心电图（ECG）是评估心律失常患者的关键工具。此外，它有助于识别患有潜在心脏病的猫，并进行心脏死亡风险分层[1]。然而，由于患者体型小、配合度可能有限，以及频繁出现寒战和呼噜声伪影，猫的小振幅偏转（如P波）分析有时可能具有挑战性[2]。
通过在左右半胸不同位置放置可变数量的电极，将胸前导联添加到传统的肢体导联系统中。已有多种胸前导联系统在人和犬中使用并验证。在人类中，胸前导联可改善心房除极、室内传导障碍和宽QRS波群心律的分析[3]。犬的研究已应用胸前导联辅助评估室性心律失常和识别室内传导障碍[4,5,6]。然而，其在犬类患者中的临床价值仍不明确。
仅有少量研究报道了胸前导联在猫中的应用，且电极放置技术各异[2,7]。最近发表的一项研究在健康猫群体中测试了九导联胸前系统[8]。迄今为止，在健康和患病猫中，将胸前导联添加到标准肢体系统的临床价值仍有待评估。
本研究旨在评估外周肢体和胸前导联系统在健康猫及患有心脏和非心脏疾病的猫群体中的可行性和临床实用性。健康猫用于生成正常描述值，并与患病猫进行比较，以检验若干ECG变量在组间的鉴别价值。
我们假设胸前导联的P波振幅高于外周肢体导联。最后，从两种系统（外周肢体和胸前）获得的ECG变量将有助于区分健康猫与患有心脏和非心脏疾病的猫。
2. 材料与方法
2.1. 研究群体
这是一项前瞻性、比较性和探索性单中心研究。2023年10月至2025年10月期间，在苏黎世大学Vetsuisse学院心脏病学分部就诊进行心脏病学评估的猫被纳入研究。未进行样本量计算，纳入动物数量基于探索性队列。
纳入标准为：猫年龄一岁或以上，完成完整的超声心动图检查，并在四分钟ECG记录期间显示窦性心律阶段。如果猫不配合，或ECG记录时存在胸腔或心包积液（量超过微量），则被排除。
对每只猫记录以下数据：信号特征、年龄、体格检查和心脏听诊结果、疾病状态，以及适用的额外诊断发现，包括实验室和非心脏影像学数据。还记录了心脏和心外疾病的存在和类型、充血性心力衰竭的存在（肺水肿和/或胸腔或腹腔积液），以及动脉血栓栓塞的存在。对于七岁以上的猫，额外评估血清甲状腺激素浓度和全身血压。使用多普勒设备（MD4，Sonomed Ltd.，华沙，波兰）测量间接收缩压。
允许轻度镇静，使用口服加巴喷丁（10–20 mg/kg）和/或肌肉注射布托啡诺（0.1–0.3 mg/kg），以减少应激并提高患者在诊断程序和ECG记录期间的配合度。不允许使用其他镇静剂。
根据诊断要求，基于临床和超声心动图数据及额外发现，将猫分为三组。如果未报告临床症状且体格和超声心动图检查正常，则判定为健康（H组）。七岁以上的猫还必须甲状腺功能亢进检测阴性，且血压值正常（<160 mmHg）。存在柔和的心脏杂音，但无超声心动图证据显示结构性心脏病，不作为排除标准。如果超声心动图检查显示任何类型的获得性或先天性心脏异常，伴或不伴心外疾病（包括甲状腺功能亢进或全身性动脉高血压），则归入心脏病（CaD）组。最后，超声心动图检查正常但存在任何类型心外疾病的猫被归类为非心脏病（NCD）组。
2.2. 超声心动图
所有猫均由委员会认证的心脏病专家或由委员会认证的心脏病专家监督下的心脏病学住院医师进行超声心动图检查。超声心动图检查按照既定兽医心脏病学标准[9]进行，使用配备不同频率相控阵探头的超声机（EPIQ 7超声系统，Philips Healthcare，Horgen，瑞士）。心脏测量值从右侧胸骨旁和左侧心尖切面的二维电影回放中获得。从乳头肌水平的短轴切面记录以下测量值：舒张末期室间隔厚度（采用前缘到后缘技术测量）、舒张末期左心室游离壁厚度（采用前缘到前缘技术测量）、左心室舒张末期和收缩末期内径（采用内缘到内缘技术测量）以及左心室缩短分数。从右侧胸骨旁短轴基底切面测量收缩末期左心房和主动脉内径，并计算其比值。从右侧胸骨旁长轴切面，采用内缘到内缘技术测量收缩末期右心房和左心房内径以及舒张末期右心室内径（三尖瓣环水平）。采用连续波多普勒从左侧心尖五腔切面记录左心室流出道最大流速。根据已发表的指南[10]，心肌病表型分类为肥厚型、限制型和非特异性型。明确的左心室肥厚定义为舒张末期存在任何左心室壁节段≥6 mm[11]，而5.5至5.9 mm之间的值被认为不确定[12]。此外，二尖瓣收缩期前向运动伴左心室壁测量值正常也被认为对肥厚型心肌病（HCM）表型不确定。如果未发现超声心动图异常，则认为猫心脏健康。此外，以下心脏测量值必须在正常范围内：舒张末期左心房和右心房内径分别<16 mm和13.72 mm[13,14]；左心房与主动脉内径比值<1.5[15]；舒张末期右心室内径（<6.36）[13]；以及舒张末期左心室壁厚度<5.5 mm[12]。
所有超声心动图研究均在患者纳入结束时由单一操作者使用专用软件（TomTec Arena，版本TTA2.51.04，TOM TEC Imaging Systems GmbH，Unterschleissheim，德国）进行审查并重复测量。
2.3. 心电图检查
使用数字心电图机（Cube ECG，Cardioline S.p.A.，Caverano，意大利）进行心电图检查，设置如下：纸速50 mm/s；灵敏度20 mm/mV；低通滤波器35 Hz；高通滤波器0.05 Hz；陷波滤波器50 Hz；采样频率500 Hz。
所有心电图记录均由一名操作者（IB或MBT）进行，持续四分钟。在猫轻度束缚于右侧卧位、腿部轻度伸展的情况下进行13导联体表心电图。使用鳄鱼夹电极直接放置在皮肤上，不剃毛，并使用酒精作为导电剂。通过将电极放置在桡骨和尺骨水平获得外周肢体导联（I、II、III、aVR、aVL、aVF）。胸前导联按照Wilson胸前导联系统记录，该系统经Kraus等人[4]修改并为本研究稍作调整，如图1所示。V1电极置于右侧第四肋间胸骨旁。V2电极置于左侧第五肋间与V1相同高度，V4电极置于与V2相同肋间但肋软骨交界水平。V3电极置于距V2和V4等距处，其余左侧胸前导联（V5和V6）置于背侧，彼此等距。记录两分钟后，将V1电极重新置于左侧第五肋间，距V6约两个电极间距，与肩肱关节相同高度（V8），再重复记录两分钟。
研究结束时，每份心电图描记均由一名对患者分组分配不知情的单一操作者（MBT）分析。所有测量均在打印的心电图上进行，使用精度为0.025 mV和10 ms的手动卡尺。
仅纳入至少显示三个连续窦性心搏的描记；在这些心搏上进行测量，并计算均值。此外，仅保留伪影最少且等电位线“干净”平直的高质量心电图。节律诊断包括窦性心律、窦性心律失常或病理性节律，并基于以下标准：正常窦性心律：心率正常，伴有规则的窦性P波，R-R间期变异小于10%；窦性心动过缓和窦性心动过速：窦性心律，心率分别低于或高于140次/分和220次/分；房性期前复合波：由形态异常的P波传导的提前出现的窄QRS复合波；室上性心动过速：连续三个以上房性期前复合波，心率高于220次/分；室性期前复合波：提前出现的宽大畸形QRS复合波，不伴相关P波[16]；加速性室性自主心律：三个以上室性期前复合波，心率正常；室性心动过速：三个以上室性期前复合波，心率高于220次/分[1,17]。心率通过计数三秒间隔内包含的QRS复合波数量并乘以二十计算。
从所有肢体导联和胸前导联测量了若干心电图变量，如图2所示。
Logarithmic formula
[内容请查看原文]
[内容请查看原文]
2.4. 统计分析
所有连续变量使用Shapiro–Wilk检验评估正态性。组间体重比较使用ANOVA检验，性别分布比较使用Fisher精确检验。组间所有其他变量比较使用非参数Kruskal–Wallis检验（H vs. CaD vs. NCD）。事后多重两两比较使用Steel–Dwass–Critchlow–Fligner程序进行。为考虑评估141个心电图变量产生的多重性，总体Kruskal–Wallis p值额外使用Benjamini–Hochberg错误发现率程序进行调整。该调整仅应用于研究范围内的单变量比较，未用作多变量逻辑分析的筛选标准，后者被视为独立的探索性分析。13个导联的P波振幅在各研究组内分别使用非参数Friedman检验对相关样本进行比较。当全局Friedman检验显著时，导联间的事后两两比较使用Conover多重比较程序进行。由于Friedman分析需要完整的重复测量，在一个或多个导联中缺失P波振幅的猫被排除在相应的组内分析之外。数据以均值±标准差或中位数（第一至第三四分位数）报告。
进行初步单变量逻辑分析以识别最显著的预测因子，包括年龄（作为连续变量引入）和镇静状态。随后的共线性分析有助于检测共线预测因子，这些因子被排除在进一步分析之外。预测因子间的共线性使用方差膨胀因子（VIF）评估。VIF值接近1表明多重共线性很小或没有，而VIF值增加表明共线性水平增加。值>5被认为提示可能存在多重共线性问题，而值>10被认为提示存在实质性多重共线性，尽管这些阈值应视为经验法则而非绝对标准[20]。最后，进行前向选择的多变量逻辑回归分析以识别最显著的预测因子。对于最终多变量模型中保留的连续预测因子，通过对相应变量进行均值中心化后，逐一将每个预测因子的二次项添加到多变量模型中来评估logit中的线性假设。统计学显著的二次项被认为是偏离线性函数形式的证据。多变量建模程序的内部验证使用2000次bootstrap重采样进行。在每个bootstrap样本中，使用相同的候选预测因子集和进入标准（p < 0.05）重复整个前向选择程序。然后将每个bootstrap样本中选择的模型应用于原始数据集。乐观度估计为bootstrap样本中获得的曲线下面积（AUC）与将相应模型应用于原始数据集时获得的AUC之间的平均差异。乐观度校正AUC通过从表观AUC中减去平均乐观度计算。还计算了各个预测因子在bootstrap样本中的选择频率以评估模型选择稳定性。
为评估镇静对心电图测量的潜在影响，使用点双列相关分析将镇静状态（0 = 未镇静；1 = 镇静）与作为候选预测因子的定量心电图变量进行比较。p值使用Benjamini–Hochberg错误发现率程序进行多重检验调整。
为探讨心脏治疗是否可能影响入组时记录的心电图表型，在CaD组内进行了额外的敏感性分析。将心电图记录时接受心脏药物治疗的猫与未治疗的CaD猫就最终多变量模型中保留的心电图变量进行比较。由于未假定这些变量呈正态分布，比较采用Mann-Whitney U检验进行。
为评估多变量研究结果的稳健性，进行了一项敏感性分析，仅考虑H组猫和CaD组中患有HCM的猫，排除其他心脏疾病和HCM诊断不明确的猫。
p值<0.05被认为具有显著性。所有统计分析使用SAS 9.4版（SAS®软件，SAS Institute Inc.：Cary，NC，USA；2013）和MedCalc 22.016版（MedCalc® Software Ltd.，Ostend，Belgium）进行。
3. 结果
3.1. 研究群体
共有168只客户拥有和工作人员拥有的猫最初纳入研究。猫被转诊至心脏科服务以明确心脏杂音、呼吸系统症状、心律紊乱、肢体跛行、虚脱、急性失明或可能与心脏病相关的全身性临床症状（呕吐、食欲减退）。其他指征包括明确在急诊床旁超声或X线检查中诊断的心脏增大或胸腔和/或腹腔游离液体。其他猫因年度心脏随访、健康筛查检查和繁殖目的的心脏检查而就诊。
其中，125只猫符合纳入标准并纳入最终分析：H组40只猫，CaD组57只猫，NCD组28只猫。
CaD组猫（9[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18]岁）和NCD组猫（10.5[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17]岁）年龄大于H组猫（4[1–15岁]）（分别为p=0.006和p=0.002），CaD与NCD之间无差异（p>0.999）。性别分布和体重在组间无差异（分别为p=0.35和p=0.507）。
共涉及18个品种。最常见的是欧洲短毛猫（n=66，52.8%），其次是英国短毛猫（n=18，14.4%）、缅因猫（n=10，8%）、孟加拉猫（n=5，4%）、挪威森林猫（n=4，3.2%）、德文卷毛猫、西伯利亚猫和苏格兰折耳猫（各n=3，2.4%）。较少涉及的品种包括波斯猫、布偶猫、塞尔凯克卷毛长毛猫（各n=2，1.6%）、伯曼猫、英国长毛猫、缅甸猫、夏特尔猫、异国长毛猫、俄罗斯蓝猫和土耳其安哥拉猫（各n=1，0.8%）。
31/125只（24.8%）猫接受了轻度镇静。21/31只猫（H组5只，CaD组14只，NCD组2只）使用布托啡诺作为单一镇静剂，剂量范围为0.1至0.3 mg/kg，肌肉注射。31只猫中有9只接受了加巴喷丁，在检查前90–120分钟给予（每组3只猫），剂量范围为10至20 mg/kg，口服给药。这9只猫中有2只（CaD组1只，NCD组1只）在诊所额外接受了一次布托啡诺联合加巴喷丁。31只猫中有1只因慢性神经病理性疼痛接受普瑞巴林（5 mg/kg q12h）。
CaD组57只猫中，33只（58%）被诊断为HCM表型。其中13只（39%）存在SAM。1只HCM表型猫被诊断有心尖薄且运动减弱节段。2只HCM表型猫出现动脉血栓栓塞（3.5%），4只猫出现充血性心力衰竭（7%）。在这些猫中，动物因不良事件稳定后记录心电图。57只猫中14只（25%）被归类为HCM不确定。其中，8/14只（57%）有不确定的左心室壁厚度，5/14只（36%）有不确定的左心室壁厚度和SAM，而1/14只（7%）仅有SAM。6/57只猫（10.5%）被诊断为非特异性心肌病表型。2/57只猫（3.5%）被诊断为限制性心肌病表型。二尖瓣狭窄和主动脉瓣下狭窄各诊断1/57只（1.7%）。
在CaD组中，27/57只猫（47%）伴有全身性疾病。全身性疾病包括甲状腺功能亢进（n=10，37%）、慢性肾病（n=6，22.2%）、全身性高血压（n=6，22.2%）、慢性肠病（n=3，11.1%）、急性胃肠炎、肾上腺肿瘤、贫血、哮喘、不明原因肺部病变（各n=2，7.4%）、膀胱结石、糖尿病、猫下泌尿道疾病、淋巴结肿大、腹部肿块、牙科疾病、肝病、胰腺炎、醛固酮增多症和维生素B缺乏（各n=1，3.7%）。一些猫同时患有不止一种疾病。在CaD组中，29只猫（50.9%）在纳入时正在接受心脏病治疗。药物包括氯吡格雷（18/29只猫）、呋塞米（8/29只猫）、利伐沙班（5/29只猫）、阿替洛尔（4/29只猫）、氨氯地平（3/29只猫）、托拉塞米（3/29只猫）、贝那普利（2/29只猫）、索他洛尔（2/29只猫）和替米沙坦（2/29只猫）。
在NCD组中，慢性肾病是最常见的诊断，慢性肠病和甲状腺功能亢进各4例（14.3%）。急性胃肠炎、慢性支气管疾病和肺部肿块各见于2只猫（7.1%）。其他诊断包括贫血、膀胱结石、糖尿病、耳道囊肿、癫痫、猫传染性腹膜炎、猫下泌尿道疾病、纤维肉瘤、肱骨骨折、甲状腺切除术后低钾血症、特发性头部震颤、肝癌、胸骨淋巴结肿大、前纵隔肿块、神经病理性疼痛、耳炎、肾盂肾炎、脓皮病、猫支原体感染导致的视网膜溃疡、球后病变、小细胞淋巴瘤、鳞状细胞癌、系统性高血压和葡萄膜炎（各1例；3.6%）。部分猫同时患有不止一种疾病。
超声心动图变量及两两比较结果汇总于附录A表A1。
3.2. 心电图检查结果
在H组（n=40）中，2/40只猫（5.0%）发现窦性心律失常，其中1只在记录期间还出现单次室性逸搏。孤立性异位搏动包括1/40只猫（2.5%）出现单个室性早搏复合波，1/40只猫（2.5%）出现房性并行心律。1/40只猫（2.5%）记录到一阵非持续性加速性室性自主心律。该猫同时伴有窦性心动过缓。
在CaD组（n=57）中，7/57只猫（12.3%）记录到室性早搏复合波，4/57只猫（7.0%）记录到房性早搏复合波。11/57只猫（19%）发现窦性心动过速。
在NCD组（n=28）中，窦性心律失常是最常见的发现，6/28只猫（21.4%）存在。其中1只猫还出现单个室性早搏复合波。5/28只猫（17.9%）观察到窦性心动过缓，其中4只同时伴有窦性心律失常。
全部13个导联各变量的描述性数值取自H组（表1）。
两两比较发现，H组与CaD组之间有35/141个心电图变量存在差异，而NCD组与H组及CaD组之间无任何变量存在差异（表1）。对所有141项总体Kruskal–Wallis检验应用了Benjamini–Hochberg错误发现率校正（附录A表A2）。校正后仍有11个变量具有显著性。错误发现率校正未用作多变量逻辑回归分析的筛选标准，该分析被视为独立的探索性分析，旨在识别心脏病的独立预测因子。单变量分析中还考虑了年龄和镇静因素，但仅年龄具有统计学显著性（p=0.002）。
在评估多重共线性后，保留27个心电图变量作为多变量逻辑回归分析的候选预测因子。由于年龄具有潜在混杂效应，另将其作为连续候选协变量纳入，最终共28个候选预测因子。由于候选预测因子存在缺失值，多变量分析在94个完整病例上进行，包括55只CaD猫和39只H猫。前向多变量逻辑回归筛选出II导联、aVL导联和V6导联的R波振幅、V8导联的P波振幅以及年龄。II导联R波振幅每增加0.05 mV，比值比（OR）为1.51（95%置信区间[CI]：1.21–1.88）；aVL导联R波振幅每增加0.05 mV，OR为1.54（95% CI：1.10–2.15）；V6导联R波振幅每增加0.05 mV，OR为1.18（95% CI：0.98–1.41）；V8导联P波振幅每增加0.05 mV，OR为2.54（95% CI：1.12–5.74）。年龄被筛选程序保留，每增加1岁的OR为1.14（95% CI：1.00–1.30）（表2）。在最终模型中，II导联和aVL导联的R波振幅以及V8导联的P波振幅仍具有显著性，而V6导联R波振幅和年龄显示边缘性关联（分别为p=0.073和p=0.053）。后两个变量在进入前向筛选程序时满足预设的显著性标准（p<0.05），但在最终模型中纳入后续预测因子后，其统计学显著性减弱。最终多变量模型中保留的任何连续预测因子均未检测到logit线性偏离的证据（所有二次项p>0.05）。由原始数据集得出的模型表观AUC为0.884。基于2000次bootstrap重采样的内部验证（每次重采样内重复整个前向筛选程序）得出平均乐观度为0.09，乐观度校正后的AUC为0.795。Bootstrap选择频率显示，II导联R波振幅在67.5%的重采样中被选中，其次是V8导联P波振幅（51.8%）、aVL导联R波振幅（42.3%）、年龄（37.2%）和V6导联R波振幅（15.7%）。这些发现表明某些预测因子的稳定性高于其他预测因子。
镇静状态与所考虑的心电图变量仅表现出弱关联。在名义显著性水平下，aVR导联P波振幅（r = −0.236，p = 0.022）、aVF导联P波振幅（r = 0.219，p = 0.034）和V4导联P波振幅（r = 0.286，p = 0.005）与镇静状态相关。然而，在27次比较中经Benjamini–Hochberg错误发现率校正后，这些关联均不再显著（所有错误发现率校正后p > 0.05）。此外，这三个心电图变量均未保留在最终多变量模型中。总体而言，这些发现未提供证据表明镇静状态与所评估的心电图测量值之间存在稳健关联。
通过将心脏病组限制为具有明确HCM表型的猫进行敏感性分析，得到更同质的72只猫群体（39只H组和33只HCM组）。前向多变量逻辑回归选择了V8导联R波振幅、II导联R波振幅、V6导联R波振幅、aVL导联R波振幅和V8导联P波振幅（附录A，表A3）。在最终模型中，V8导联P波振幅（p = 0.012）、II导联R波振幅（p < 0.001）和aVL导联R波振幅（p = 0.003）仍与HCM显著相关，而V6导联R波振幅显示临界关联（p = 0.065），V8导联R波振幅不显著（p = 0.551）。模型整体显著（似然比p < 0.0001），并显示出高表观判别力（AUC = 0.944）。值得注意的是，在主要H对CaD分析中选择的五个变量中有四个也在HCM限制性分析中被选择（II导联、aVL导联和V6导联R波振幅以及V8导联P波振幅），其中三个在最终敏感性模型中保持统计学显著性。
入组时，CaD组57只猫中有29只正在接受心脏药物治疗，而28只未治疗。在仅限于CaD猫的敏感性分析中，保留在最终多变量模型中的心电图变量在接受治疗与未治疗猫之间均无显著差异。这些发现未提供证据表明治疗状态与多变量分析确定的主要心电图变量之间存在重大关联。
在各组中，几乎所有13个导联的P波振幅在几乎所有两两比较中均存在显著差异（表3、表4和表5；图3）。外周肢体导联的P波振幅始终高于胸前导联。然而，在H组和NCD组中，I导联P波振幅通常与胸前导联无差异（p = 0.398–0.991），而在CaD组中，其通常显示低于V5–V6导联的数值。
评估了三组之间胸前导联的R/S比值（附录A，表A4）。在V1导联中，所有组的R/S比值主要<1（H组57.5%，CaD组69.1%，NCD组62.5%）。在左侧胸前导联（V2–V8）中，所有组的R/S主要>1。
4. 讨论
猫的13导联心电图系统在区分H猫与CaD猫方面具有中等准确性，但不能区分NCD猫。然而，与我们的初始假设相反，外周肢体导联的绝对P波振幅高于胸前导联，而在胸前导联中，V1导联P波振幅的绝对值最低。添加V8导联可能不会改善窦性P波的检测，因为在三组中均未观察到与其他左侧胸前导联的差异。
在人类中，标准12导联心电图包括六个单极胸前导联（V1–V6），分布在胸部前壁和侧壁[3]。V1和V2电极分别放置在右侧和左侧胸骨旁第四肋间，V3位于V2和V4之间，V4位于锁骨中线第五肋间。V5和V6位于V4的水平面上，分别位于腋前线和中腋线。已引入额外的后壁导联（V7–V9）以改善对左心室区域电活动的检测，这些区域在常规六个胸前导联中代表性不足，特别是下基底壁和后侧壁。其中，V8放置在肩胛骨尖端，与V6在同一水平面上[3]。在人类中，V7–V9增加了胸部的空间覆盖范围，特别是在下基底壁和后侧壁左心室区域[21,22]。类似地，我们方案中采用的V8位置旨在将猫胸部的覆盖范围扩展到常规六个胸前导联之外，并优化朝向左侧背侧胸廓投影的心房信号的记录角度。因此，V8电极放置在左侧第五肋间，肩肱关节水平。
多项研究已为猫外周肢体导联建立了参考值，并描述了体位对记录波形的影响[23,24,25]。近期一项研究[8]探讨了选定心前区导联（V1、V2和V4）在健康猫中的应用价值，与我们研究类似，作者观察到心前区导联的P波振幅较低，其中V1记录的数值最低。在犬中，一些心前区导联系统也被用于更好地识别P波。在一项研究中，Santilli等[5]表明，将V1重新放置于右侧第一肋间可更可靠地检测不同胸廓形态下的右心房活动。Kurogochi等[26]测试了一套扩展心前区导联系统，包括后部导联V10和两个双极导联（M6M1和M6M2）。在这些导联上，P波振幅似乎高于常规肢体导联和心前区导联，类似于人类患者中采用的Lewis导联系统[27]。在人类中，合成的18导联心电图通过背部导联V7–V9扩展了标准系统，且V7中的P波形态已被证明可区分起源于解剖上相邻结构的心房心律失常[28]。然而，根据我们的结果，猫窦性P波的电向量似乎并不与该方向一致；因此，在猫中，无论是否患有心脏病，左侧心前区导联均不能改善对这些低电压事件的检测。
在H组评估的心电图变量中，QRS波群持续时间与既往文献报道的参考值不同。事实上，尽管健康猫的正常上限通常报道为<40 ms[2,16,19]，但在我们的40只健康猫队列中，II导联QRS持续时间的第三四分位数为60 ms。虽然我们的对照群体规模不大，但足以提供可靠的描述性数值。此外，所有猫均接受了超声心动图检查以确认无结构性心脏病，且所有心电图测量均由一位对动物临床状态不知情的有经验心脏病专家完成。观察到的QRS持续时间较长的一个可能解释是当代数字心电图系统质量的提高。更高的信号分辨率以及所用的记录设置可能有助于识别低振幅偏转，尤其是Q波和S波，从而导致测得的QRS持续时间较早期研究报告的略长。然而，不能完全排除少数猫存在亚临床室内传导障碍的可能性，这可能导致了我们队列中QRS持续时间的轻度延长。未来研究应考虑，猫的高分辨率心电图记录可能产生比既往报道更长的测得QRS持续时间。在解读QRS异常（包括宽QRS波群心律和束支传导阻滞）时应考虑到这一点。
在H组与CaD组之间，多项心电图变量存在差异，而没有任何变量能够区分H组与NCD猫。在成对比较中H组与CaD组存在差异的35个变量中，四个心电图变量和年龄在多变量模型中被保留为CaD的预测因子。这些变量分别是II导联R波振幅、aVL导联R波振幅、V6导联R波振幅和V8导联P波振幅。然而，年龄和V6导联R波振幅处于临界显著性水平，因此可能相关性较低；当仅纳入具有HCM表型的猫进行亚组分析时，它们未被保留为显著变量。在最近的一项研究中[29]，II导联R波振幅在健康猫与代偿期（临床前期）HCM猫之间显示出最高的判别价值，截断值为>0.26 mV。然而，在Gardner等人的研究[29]中，该变量的准确性相对较低，因为患病组的所有猫均处于疾病的亚临床阶段。在我们的研究中，II导联R波振幅每增加0.05 mV，与患CaD的几率增加1.54倍及具有HCM表型的几率增加1.9倍相关。在人类中，aVL导联R波振幅增高被认为是左心室肥厚Cornell电压标准的组成部分[30]，并且与一般人群中升高的心血管发病率和死亡率相关[31]。类似的解释可能适用于猫的结构性心脏病，其中左心室侧壁增厚可使电压向aVL导联偏移。尽管我们的研究并不限于左心室肥厚的猫，但HCM是最常见的心脏异常类型，占该组57只患猫中的33只，而14只猫为可疑HCM。事实上，CaD组的左心室最大壁厚度值高于其他两组。此外，当亚组分析仅考虑HCM猫时，aVL导联R波振幅仍具有显著性。因此，aVL导联R波振幅增高可能如同在人类中一样，可预测猫的左心室肥厚。同样，心前导联V6和V8的R波振幅在预测CaD和HCM方面处于临界显著性水平。这支持了R波振幅可能反映左心室壁肥厚存在的证据，正如在其他研究中所观察到的[29]。
V8导联P波振幅在多变量分析中被确定为CaD的另一预测因子。我们假设这反映了猫心肌病中左心房增大的高患病率[14]，并支持左侧心前导联对心房活动评估的贡献。这一观察与Santilli等人[8]的发现一致，他们报道健康猫左侧心前导联的P波振幅高于右侧心前导联。由于H组猫比患病猫年轻，年龄在多变量分析中作为连续变量进行分析，结果发现年龄与上述其他心电图变量一样，是CaD和HCM的危险因素。这一发现已在先前的研究中有所报道，显示年龄较大的猫发生HCM的风险更高，并强调了异常表型表达可能随年龄增长而增加，如同在人类中一样[32]。
在我们的人群中，H组与CaD组之间的QT和QTc间期无显著差异。这与Romito等人[1]的发现相反，他们报道左心室肥厚猫的六导联心电图上QT和QTc延长。同样，Bastos等人[33]记录了具有HCM表型的猫与健康对照相比QT间期更长且QT不稳定性增加。这种差异的可能原因包括我们CaD组的异质性。CaD组的大多数猫患有心肌病，涉及多种表型，而较小比例的猫表现为瓣膜性或其他心脏疾病。此外，猫纳入时处于不同的疾病阶段。相比之下，Romito等人[1]和Bastos等人[33]专门纳入了具有肥厚型表型的猫。
我们关于心前导联R/S比值的发现与Santilli等人[8]所描述的相当。各组中V1导联的R/S比值主要<1，而在左侧心前导联（V3–V8）中，大多数猫的R/S比值>1。在各组之间检测到了主要的R/S模式；CaD组中V3–V8导联R/S比值偏高的趋势与左心室肥厚时QRS向量向左和向背侧偏移相一致[19]。
应承认若干局限性。首先，包括患病猫在内的各组显示出多种不同的心脏和非心脏疾病。此外，CaD组中的一些猫还患有非心脏性和全身性疾病。然而，我们认为这更能代表日常实践中典型的真实临床情况。主要目的是评估心电图在健康猫和患病猫之间的鉴别价值，并评估非心脏疾病是否可能影响心电图变量。为克服这一局限性，我们进行了一项亚组分析，仅比较H组猫和CaD组中表现为HCM表型的猫。结果与原始分析相似，几乎保留了所有相同的变量作为疾病的显著预测因子（即II、aVL、V6和V8导联的R波振幅以及V8导联的P波振幅）。
其次，允许对部分猫进行轻度镇静，以减少应激并提高检查质量。近期在猫中进行的随机研究表明，单次口服加巴喷丁仅引起心血管参数的轻微变化[34,35]，且对标准心电图变量无显著影响[36]。布托啡诺是本研究中唯一使用的阿片类镇静剂，而未允许使用阿法沙酮。在一项健康猫的研究中，肌内联合给予阿法沙酮和布托啡诺未产生临床相关的超声心动图测量变化，但未评估其对心电图的影响[37]。据我们所知，尚无研究专门评估单独使用布托啡诺或布托啡诺联合加巴喷丁对猫心电图变量的影响。我们的分析显示，镇静状态与任何心电图变量之间均不存在关联。
CaD组中的一些猫正在接受药物（如利尿剂和抗心律失常药）治疗，这些药物可能影响某些心电图变量，如振幅或心率。尽管在敏感性分析中，多变量分析确定的主要心电图变量在接受治疗和未接受治疗的CaD猫之间无差异，但不能排除治疗效应。治疗并非随机分配，而是根据临床指征开具；因此，接受治疗和未接受治疗的猫之间的差异也可能反映疾病严重程度和其他临床特征。因此，本研究无法完全区分疾病相关的心电图变化与治疗相关效应。
第三，尽管电极仅由两名操作者按照标准化方案放置，但不能排除肢体或胸前导联位置的一些微小误差。焦虑的猫前肢可能未能最佳伸展，超重猫的肋间隙和胸骨可能难以触及，这些猫可能存在电极位置不正确的情况。此外，为记录13个导联，V1电极被替换至V8位置。因此，无法同时记录所有13个导联。在两个记录阶段期间，心率或应激状态的轻微变化可能影响了结果。
第四，该群体是在单一转诊心脏病学服务机构招募的，使用一台心电图机及特定设置，因此不一定代表一般猫群体。特别是，500 Hz的采样频率虽在文献中被广泛采用，但可能不足以准确分辨猫的心电图波形，因为猫特征性的心率升高需要更高的时间分辨率。
第五，本研究仅纳入窦性心律的猫。我们的目的并非评估心电图诊断任何特定心律失常的准确性，也非评估胸前导联在识别非窦性P波中的辅助价值。确实，在某些情况下，当P'波在外周肢体导联中不可见时，胸前导联可能有助于显现这些低电压事件，尤其是当P'波电轴显著偏移时，如房性心动过速或旁路介导或交界性心律期间的逆行P'波。
最后，关于多变量模型，从原始数据集得出的模型的表观区分度较高（AUC=0.884）。然而，由于预测变量选择和模型估计是在同一队列中进行的，该值预期会高估模型性能。因此，通过在2000个bootstrap样本中各自重复整个变量选择流程，进行了内部bootstrap验证。经乐观度校正后，AUC降至0.795。因此，尽管部分表观性能归因于在同一数据集中进行模型开发，但建模流程在内部验证后仍保持了可接受至良好的区分能力。Bootstrap选择频率也显示模型的组成并不完全稳定。II导联R波振幅显示出最大的稳定性，在约三分之二的bootstrap样本中被选中，而V8导联P波振幅在约半数样本中被保留。其余变量的选择一致性较低，尤其是V6导联R波振幅。这种变异性可能反映了相对有限的样本量以及多项心电图测量提供部分重叠信息的情况。因此，最终模型中保留的各个变量不应被视为心脏病的确定性预测因子，而应被视为在该队列中与心脏病相关的变量，其相关性需要在独立人群中加以确认。仅限于具有明确HCM表型的猫进行的敏感性分析为主要多变量发现提供了额外支持。尽管分析仅限于一个较小但临床上更为同质的心脏病群体，主要分析中选出的五个变量中有四个被前向选择流程保留。特别是，II导联、aVL导联和V6导联的R波振幅以及V8导联的P波振幅在最终的HCM限制模型中仍与疾病状态显著相关，V8导联R波振幅作为额外的显著变量。这种一致性表明，这些心电图测量所观察到的关联并非仅由整体CaD组中所包含的心脏异常异质性所驱动。
5. 结论
总之，包含外周肢体导联和七个胸前导联的13导联心电图系统可应用于猫。外周肢体导联在所有导联中具有最高的P波振幅。然而，V8导联提供的P波振幅高于V1导联，可能有助于评估心房电活动。本研究还提供了健康成年猫外周肢体导联的描述性数值，并作为新贡献提供了七个胸前导联的描述性数值。一些胸前导联和外周肢体导联变量，特别是II导联和aVL导联的R波振幅以及V8导联的P波振幅，是CaD的预测因子，而NCD猫与H猫之间无任何心电图参数存在差异。此外，相同的变量有助于区分H猫与具有HCM表型的猫。
查看英文原文▾
1. Introduction
Surface electrocardiogram (ECG) is a key tool for evaluating patients with arrhythmia. Moreover, it can help identifying cats with underlying cardiac disease and stratify for risk of cardiac death [1]. However, due to the small patient size, possible limited cooperation, and frequent artifacts from shivering and purring, the analysis of small amplitude deflections, like P waves, might sometimes be challenging in cats [2].
Precordial leads are added to the traditional peripheral limb system by placing a variable number of electrodes at different points on the right and left hemithorax. Several precordial lead systems have been used and validated in people and dogs. In humans, precordial leads improve analysis of atrial depolarizations, intraventricular conduction disturbances, and wide QRS complex rhythms [3]. Studies in dogs have applied precordial leads to aid evaluating ventricular arrhythmias and identify intraventricular conduction disturbances [4,5,6]. However, their clinical value in canine patients remains unclear.
Only a small number of studies report the use of precordial leads in cats, with various techniques of electrode application [2,7]. A recently published study tested a nine-lead precordial system in a population of healthy cats [8]. To date, the clinical value of adding precordial leads to the standard limb system in healthy and diseased cats remains to be evaluated.
This study aims to assess the feasibility and clinical utility of a peripheral limb and precordial lead system in a population of healthy cats and cats with cardiac and non-cardiac diseases. The healthy cats are used to generate normal descriptive values and compared to diseased cats in order to test the discriminatory value of several ECG variables between groups.
We hypothesized that P wave amplitudes are higher in precordial leads compared to peripheral limb leads. Finally, ECG variables obtained from both systems (peripheral limb and precordial) will allow differentiating between healthy cats and cats with cardiac and non-cardiac diseases.
2. Materials and Methods
2.1. Study Population
This was a prospective, comparative, and explorative single-center study. Between October 2023 and October 2025, cats presented to the Division of Cardiology of the Vetsuisse Faculty at the University of Zurich for cardiological assessment were included. No sample-size calculation was performed, and the number of animals included was based on an exploratory cohort.
To be included, cats had to be one year of age or older, have a complete echocardiographic examination performed, and show phases of sinus rhythm during a four-minute ECG recording. Cats were excluded if not cooperative or if pleural or pericardial effusion (in a quantity more than traces) was present at the time of ECG recording.
For each cat, the following data were annotated: signalment, age, physical examination and cardiac auscultation findings, disease status, and additional diagnostic findings when applicable, including laboratory and non-cardiac imaging data. Presence and type of cardiac and extra-cardiac disease, presence of congestive heart failure (lung edema, and/or pleural or abdominal effusion), and presence of arterial thromboembolism were also annotated. In cats older than seven years, serum thyroid hormone concentration and systemic blood pressure were additionally assessed. Indirect systolic blood pressure was measured using a Doppler device (MD4, Sonomed Ltd., Warsaw, Poland).
Mild sedation was allowed with oral gabapentin (10–20 mg/kg) and/or intramuscular injection of butorphanol (0.1–0.3 mg/kg) to reduce stress and improve patients’ cooperation during diagnostic procedures and ECG recording. No other sedative was allowed.
Cats were divided into three groups based on clinical and echocardiographic data and additional findings according to the diagnostic requirements. Cats were deemed to be healthy (H group) if no clinical signs were reported and if physical and echocardiographic examinations were normal. Cats older than seven years had also to test negative for hyperthyroidism and had to have normal blood pressure values (<160 mmHg). Presence of a soft heart murmur, in absence of echocardiographic evidence of structural heart disease, was not an exclusion criterion. Cats were allocated to the cardiac disease (CaD) group if echocardiographic exam showed any type of acquired or congenital cardiac abnormality, with or without concomitant extra-cardiac diseases, including hyperthyroidism or systemic arterial hypertension. Finally, cats with a normal echocardiographic examination and presence of any type of extra-cardiac disease were classified in the non-cardiac disease (NCD) group.
2.2. Echocardiography
All cats underwent echocardiographic examination performed by a board-certified cardiologist or a cardiology resident in training supervised by a board-certified cardiologist. Echocardiographic examinations were conducted according to the established veterinary cardiology standards [9] using an ultrasound machine equipped with phased array transducers of different frequencies (EPIQ 7 ultrasound system, Philips Healthcare, Horgen, Switzerland). Cardiac measurements were obtained from two-dimensional cine loops from right parasternal and left apical views. From a short-axis view, at the level of the papillary muscles, the following measurements were recorded: end-diastolic interventricular septum thickness (measured using a leading edge-to-trailing edge technique), end-diastolic left ventricular free wall thickness (measured using a leading edge-to-leading edge technique), left ventricular end-diastolic and end-systolic internal diameters (measured using an inner edge-to-inner edge technique) and left ventricular fractional shortening. From a right parasternal short-axis basilar view, the left atrial and aortic internal diameters at end-systole were measured, and their ratio calculated. From a right parasternal long-axis view, the internal diameter of the right atrium and left atrium at end-systole and of the right ventricle and end-diastole (at the tricuspid annulus level) were measured using an inner edge-to-inner edge technique. Left ventricular outflow tract maximal velocity was recorded using a continuous wave Doppler from a left apical five-chambers view. Cardiomyopathy phenotypes were classified as hypertrophic, restrictive and non-specific following published guidelines [10]. Clear left ventricular hypertrophy was defined as the presence of any left ventricular wall segments ≥6 mm at end-diastole [11], while values between 5.5 and 5.9 mm were considered equivocal [12]. Moreover, the presence of systolic anterior motion of the mitral valve in combination with normal left ventricular wall measurements was also considered equivocal for a hypertrophic cardiomyopathy (HCM) phenotype. Cats were considered cardiologically healthy if no echocardiographic abnormalities were identified. Moreover, the following cardiac measurements had to be within normal limits: left atrial and right atrial diameter at end-diastole <16 mm and 13.72 mm, respectively [13,14]; ratio between the left atrial and aortic diameter < 1.5 [15]; right ventricular internal diameters at end-diastole (<6.36) [13]; and left ventricular wall thickness at end-diastole <5.5 mm [12].
All echocardiographic studies were reviewed, and measurements were repeated by one single operator at the end of the patients’ inclusion using dedicated software (TomTec Arena, version TTA2.51.04, TOMTEC Imaging Systems GmbH, Unterschleissheim, Germany).
2.3. Electrocardiography
Electrocardiographic examinations were obtained using a digital ECG machine (Cube ECG, Cardioline S.p.A., Caverano, Italy) using the following settings: paper speed, 50 mm/s; sensitivity, 20 mm/mV; low-pass filter, 35 Hz; high-pass filter, 0.05 Hz; notch filter, 50 Hz; sampling frequency, 500 Hz.
All the ECG recordings were performed for a duration of four minutes by one operator (IB or MBT). A 13-lead surface ECG was performed with the cats mildly restrained in right lateral recumbency, with the legs slightly stretched. Alligator electrodes were employed and directly placed on the skin, without clipping the hair and using alcohol as an electrical conductor. Peripheral limb leads (I, II, III, aVR, aVL, aVF) were obtained by placing the electrodes at the level of the radius and ulna. Precordial leads were recorded by positioning the electrodes according to Wilson’s precordial lead system, as modified by Kraus et al. [4] and slightly adjusted for this study, as shown in Figure 1. The V1 electrode was positioned at the fourth right intercostal space parasternally. The V2 electrode was positioned at the fifth left intercostal space at the same height as V1, while the V4 electrode was placed at the same intercostal space as V2 but at the level of the costochondral junction. The V3 electrode was placed equidistant from V2 and V4, and the remaining left precordial electrodes (V5 and V6) were placed dorsally, equidistant from each other. After two minutes of recording, the V1 electrode was repositioned at the fifth left intercostal space, approximately two interelectrode distances from V6, at the same height as the scapulohumeral joint (V8), and the recording was repeated for two additional minutes.
Each ECG trace was analyzed at the end of the study by a single operator (MBT) blinded to the patient’s group allocation. All measurements were performed on printed ECG, using a manual caliper with a precision of 0.025 mV and 10 ms.
Only traces showing at least three consecutive sinus beats were included; measurements were performed on those beats, and the mean was calculated. Moreover, only good-quality ECGs with minimal artifacts and “clean” and straight isoelectric lines were retained. Rhythm diagnoses included sinus rhythm, sinus arrhythmia or pathological rhythms and were based on the following criteria; normal sinus rhythm: normal heart rate coupled to regular sinus P waves, with less than 10% variation in R-R intervals; sinus bradycardia and sinus tachycardia: sinus rhythm with heart rate below or above 140 and 220 beats per minute, respectively; atrial premature complex: a premature narrow QRS complex conducted by a P wave with an abnormal morphology; supraventricular tachycardia: more than three consecutive atrial premature complexes with a heart rate above 220 beats per minute; ventricular premature complex: premature wide and bizarre QRS complex without an associated P wave [16]; accelerated idioventricular rhythm: more than three ventricular premature complexes at a normal heart rate; ventricular tachycardia: more than three ventricular premature complexes at a heart rate above 220 beats per minute [1,17]. Heart rate was calculated by counting the number of QRS complexes included within a three-second interval and multiplied by twenty.
Several ECG variables were measured from all limb and precordial leads, as shown in Figure 2.
Logarithmic formula
[内容请查看原文]
[18] was used to correct the QT interval in lead II for the heart rate (QTc). Although this equation has not been validated in cats, it showed the greatest accuracy in calculating the canine QTc compared with other methods, and it has also been used in other feline studies [1,18]. The mean electrical axis on the frontal plane for P and T waves and QRS complex was calculated using the following formula[内容请查看原文]
[19].2.4. Statistical Analysis
All continuous variables were assessed for normality using the Shapiro–Wilk test. Body weight among groups was compared using an ANOVA test, while sex distribution was compared with a Fisher’s exact test. All other variables among groups were compared using a non-parametric Kruskal–Wallis test (H vs. CaD vs. NCD). Post hoc multiple pairwise comparisons were performed using the Steel–Dwass–Critchlow–Fligner procedure. To account for the multiplicity arising from the evaluation of 141 ECG variables, the overall Kruskal–Wallis p-values were additionally adjusted using the Benjamini–Hochberg false discovery rate procedure. This adjustment was applied only to the study-wide univariate comparisons and was not used as a screening criterion for the multivariate logistic analysis, which was considered a separate exploratory analysis. P wave amplitudes across the 13 leads were compared separately within each study group using the non-parametric Friedman test for related samples. When the global Friedman test was significant, post hoc pairwise comparisons between leads were performed using Conover’s multiple-comparison procedure. Because the Friedman analysis requires complete repeated measurements, cats with a missing P wave amplitude in one or more of the leads were excluded from the corresponding within-group analysis. Data are reported as mean ± standard deviation or median (first–third quartile).
Preliminary univariate logistic analyses were performed to identify the most significant predictors, including age (introduced as a continuous variable) and sedation status. A subsequent collinearity analysis helped to detect collinear predictors, which were excluded from further analyses. Collinearity among predictors was assessed using the variance inflation factor (VIF). VIF values close to 1 indicate little or no multicollinearity, whereas increasing VIF values indicate increasing levels of collinearity. Values > 5 were considered indicative of potentially problematic multicollinearity, while values > 10 were considered indicative of substantial multicollinearity, although these thresholds should be regarded as rules of thumb rather than absolute criteria [20]. Finally, a multivariate logistic regression analysis with forward selection was conducted to identify the most significant predictors. For the continuous predictors retained in the final multivariable model, the assumption of linearity in the logit was assessed by adding a quadratic term for each predictor to the multivariable model, one at a time, after mean-centering the corresponding variable. A statistically significant quadratic term was considered evidence of departure from a linear functional form. Internal validation of the multivariable modelling procedure was performed using 2000 bootstrap resamples. In each bootstrap sample, the entire forward-selection procedure was repeated using the same candidate predictor set and entry criterion (p < 0.05). The model selected in each bootstrap sample was then applied to the original dataset. Optimism was estimated as the mean difference between the area under the curve (AUC) obtained in the bootstrap sample and that obtained when the corresponding model was applied to the original dataset. The optimism-corrected AUC was calculated by subtracting the mean optimism from the apparent AUC. Selection frequencies of individual predictors across bootstrap samples were also calculated to assess model-selection stability.
To evaluate the potential influence of sedation on ECG measurements, sedation status (0 = non-sedated; 1 = sedated) was compared with the quantitative ECG variables considered as candidate predictors using point-biserial correlation analysis. p-values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate procedure.
To explore whether cardiac treatment may have influenced the ECG phenotype recorded at enrollment, an additional sensitivity analysis was performed within the CaD group. Cats receiving cardiac medication at the time of ECG recording were compared with untreated CaD cats for the ECG variables retained in the final multivariable model. Because these variables were not assumed to be normally distributed, comparisons were performed using the Mann–Whitney U test.
To evaluate the robustness of the multivariable findings, a sensitivity analysis was performed only considering H cats and cats from the CaD group with HCM and excluding other cardiac diseases and cats with an equivocal diagnosis of HCM.
p-values < 0.05 were considered significant. All statistical analyses were performed using SAS version 9.4 (SAS® software, SAS Institute Inc.: Cary, NC, USA; 2013) and MedCalc version 22.016 (MedCalc® Software Ltd., Ostend, Belgium).
3. Results
3.1. Study Population
A total of 168 client-owned and staff-owned cats were initially enrolled in the study. Cats were presented to the cardiology service for clarification of heart murmur, respiratory complaints, rhythm disturbances, limb lameness, collapse, acute blindness, or systemic clinical signs potentially associated with heart disease (vomiting, hyporexia). Additional indications included clarification of cardiomegaly or free fluids in the thoracic and/or abdominal cavity diagnosed during emergency bedside ultrasound or radiography. Other cats were presented for annual cardiac follow-up, health screening examinations, and cardiological examination for breeding purposes.
Of these, 125 cats met the inclusion criteria and were included in the final analysis: 40 cats in the H group, 57 in the CaD group, and 28 in the NCD group.
Cats in CaD group (9 [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18] years) and NCD group (10.5 [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17] years) were older than cats in H group (4 [1–15 years]) (p = 0.006 and p = 0.002, respectively), with no difference between CaD and NCD (p > 0.999). Sex distribution and body weight did not differ between groups (p = 0.35 and p = 0.507, respectively).
In total, 18 breeds were represented. The most common was European shorthair (n = 66, 52.8%), followed by British shorthair (n = 18, 14.4%), Maine coon (n = 10, 8%), Bengal (n = 5, 4%), Norwegian forest cat (n = 4, 3.2%), Devon rex, Siberian cat, and Scottish fold (n = 3 each, 2.4%). Less frequently represented breeds included Persian cat, Ragdoll, Selkirk rex longhair (n = 2 each, 1.6%), Birman, British longhair, Burma, Chartreux, Exotic longhair, Russian blue and Turkish Angora cat (n = 1 each, 0.8%).
Mild sedation was administered to 31/125 (24.8%) cats. Butorphanol was used as single sedative in 21/31 cats (5 in the H group, 14 in the CaD group, and 2 in the NCD group), with a dose ranging from 0.1 to 0.3 mg/kg and given intramuscularly. Nine of thirty-one cats received gabapentin, given 90–120 min prior to examination (three cats in each group), with a dose ranging from 10 to 20 mg/kg, administered orally. Two of these nine cats, one cat in the CaD group and one cat in the NCD group, received butorphanol additionally to gabapentin once in the clinic. One of thirty-one cats received pregabalin (5 mg/kg q12 h) for chronic neuropathic pain.
Of the 57 cats in group CaD, 33 (58%) were diagnosed with HCM phenotype. Of these cats 13 (39%) had SAM. One cat with HCM phenotype was diagnosed with an apical thin and hypokinetic segment. Two cats with an HCM phenotype presented with arterial thromboembolism (3.5%), and four cats with congestive heart failure (7%). In these cats, ECG was recorded once the animals were stabilized for their adverse event. Fourteen of fifty-seven (25%) cats were classified as equivocal for HCM. Of these, 8/14 (57%) had an equivocal left ventricular wall thickness, 5/14 (36%) had an equivocal left ventricular wall thickness and SAM, while 1/14 (7%) had SAM only. A non-specific cardiomyopathy phenotype was diagnosed in 6/57 cats (10.5%). Restrictive cardiomyopathy phenotype was diagnosed in 2/57 cats (3.5%). Mitral valve stenosis and subaortic stenosis were each diagnosed in 1/57 (1.7%).
In the CaD group, 27/57 cats (47%) had a concomitant systemic disease. Systemic diseases included hyperthyroidism (n = 10, 37%), chronic kidney disease (n = 6, 22.2%), systemic hypertension (n = 6, 22.2%), chronic enteropathy (n = 3, 11.1%), acute gastroenteritis, adrenal gland tumor, anemia, asthma, lung pathology of unknown origin (n = 2 each, 7.4%), cystolithiasis, diabetes mellitus, feline lower urinary tract disease, lymphadenomegaly, abdominal mass, dental complaints, hepatopathy, pancreatitis, hyperaldosteronism, and vitamin B deficiency (n = 1 each, 3.7%). Some cats had more than one disease at the same time. In the CaD group, 29 cats (50.9%) were receiving treatment for their cardiac disease at the time of inclusion. Medications included clopidogrel (18/29 cats), furosemide (8/29 cats), rivaroxaban (5/29 cats), atenolol (4/29 cats), amlodipine (3/29 cats), torasemide (3/29 cats), benazepril (2/29 cats), sotalol (2/29 cats), and telmisartan (2/29 cats).
Chronic kidney disease was the most common diagnosis in the NCD group together with chronic enteropathy and hyperthyroidism (n = 4 each, 14.3%). Acute gastroenteritis, chronic bronchial disease, and lung masses were found in two cats each (7.1%). Other diagnoses included anemia, cystolithiasis, diabetes mellitus, ear canal cyst, epilepsy, feline infectious peritonitis, feline lower urinary tract disease, fibrosarcoma, humerus fracture, hypokalemia post thyroidectomy, idiopathic head tremor, liver carcinoma, lymphadenomegaly of sternal lymph nodes, a mass in the cranial mediastinum, neuropathic pain, otitis, pyelonephritis, pyoderma, retinal ulcer due to Mycoplasma felis infection, retrobulbar lesion, small cell lymphoma, squamous cell carcinoma, systemic hypertension and uveitis (n = 1 each; 3.6%). Some cats had ore that one disease at the same time.
Echocardiographic variables and results of the pairwise comparison are summarized in Appendix A, Table A1.
3.2. Electrocardiographic Findings
In the H group (n = 40), sinus arrhythmia was identified in 2/40 cats (5.0%), one of which also presented a single ventricular escape beat during the recording. Isolated ectopy comprised a single ventricular premature complex in 1/40 (2.5%), and atrial parasystole in 1/40 (2.5%). A run of non-sustained accelerated idioventricular rhythm was recorded in 1/40 (2.5%). This cat had a concurrent sinus bradycardia.
In the CaD group (n = 57), ventricular premature complexes were recorded in 7/57 cats (12.3%), and atrial premature complexes in 4/57 cats (7.0%). Sinus tachycardia was identified in 11/57 cats (19%).
In the NCD group (n = 28), sinus arrhythmia was the most common finding, identified in 6/28 cats (21.4%). One of these cats also showed a single ventricular premature complex. Sinus bradycardia was observed in 5/28 cats (17.9%), four of which had concurrent sinus arrhythmia.
Descriptive values for each variable for all 13 leads were obtained from the H group (Table 1).
Pairwise comparison identified 35/141 ECG variables as different between H and CaD groups, while no variable differed between the NCD and H and CaD groups (Table 1). The Benjamini–Hochberg false discovery rate correction was applied to all 141 overall Kruskal–Wallis tests (Appendix A, Table A2). Eleven variables remained significant after adjustment. The false discovery rate adjustment was not used as a screening criterion for the multivariate logistic analysis, which was considered a separate exploratory analysis aimed at identifying independent predictors of cardiac disease. Age and sedation were also considered in the univariate analysis, but only age remained statistically significant (p = 0.002).
Following the assessment of multicollinearity, 27 ECG variables were retained as candidate predictors for the multivariable logistic regression analysis. Age was additionally included as a continuous candidate covariate because of its potential confounding effect, resulting in a total of 28 candidate predictors. Due to missing values across the candidate predictors, the multivariable analysis was performed on 94 complete cases, comprising 55 cats with CaD and 39 H cats. Forward multivariable logistic regression selected R wave amplitude in lead II, aVL, and V6, P wave amplitude in V8, and age. A 0.05 mV increase in R wave amplitude in lead II was associated with an odds ratio (OR) of 1.51 (95% confidence interval [CI], 1.21–1.88); a 0.05 mV increase in R wave amplitude in aVL with an OR of 1.54 (95% CI, 1.10–2.15); a 0.05 mV increase in R wave amplitude in V6 with an OR of 1.18 (95% CI, 0.98–1.41); and a 0.05 mV increase in P wave amplitude in V8 with an OR of 2.54 (95% CI, 1.12–5.74). Age was retained by the selection procedure, with an OR of 1.14 per one-year increase (95% CI, 1.00–1.30) (Table 2). In the final model, R wave amplitude in lead II and aVL and P wave amplitude in V8 remained significant, whereas R wave amplitude in V6 and age showed borderline associations (p = 0.073 and p = 0.053, respectively). The latter two variables had met the predefined significance criterion at the step at which they entered the forward-selection procedure (p < 0.05), but their statistical significance was attenuated after inclusion of the subsequent predictors in the final model. No evidence of departure from linearity in the logit was detected for any of the continuous predictors retained in the final multivariable model (all quadratic terms p > 0.05). The model derived from the original dataset had an apparent AUC of 0.884. Internal validation based on 2000 bootstrap resamples, with repetition of the entire forward-selection procedure within each resample, yielded a mean optimism of 0.09 and an optimism-corrected AUC of 0.795. Bootstrap selection frequencies showed that R wave amplitude in lead II was selected in 67.5% of resamples, followed by P wave amplitude in V8 (51.8%), R wave amplitude in aVL (42.3%), age (37.2%), and R wave amplitude in V6 (15.7%). These findings indicated greater stability for some predictors than for others.
Sedation status showed only weak associations with the ECG variables considered. At the nominal significance level, P wave amplitude in aVR (r = −0.236, p = 0.022), in aVF (r = 0.219, p = 0.034), and in V4 (r =0.286, p = 0.005) was associated with sedation status. However, none of these associations remained significant after Benjamini–Hochberg false discovery rate correction across the 27 comparisons (all false discovery rate-adjusted p > 0.05). Moreover, none of these three ECG variables was retained in the final multivariable model. Overall, these findings provided no evidence of a robust association between sedation status and the ECG measurements evaluated.
A sensitivity analysis was performed by restricting the cardiac disease group to cats with a definite HCM phenotype, resulting in a more homogeneous population of 72 cats (39 H and 33 HCM). Forward multivariable logistic regression selected R wave amplitude in V8, R wave amplitude in lead II, R wave amplitude in V6, R wave amplitude in aVL, and P wave amplitude in V8 (Appendix A, Table A3). In the final model, P wave amplitude in V8 (p = 0.012), R wave amplitude in lead II (p < 0.001), and R wave amplitude in aVL (p = 0.003) remained significantly associated with HCM, whereas R wave amplitude in V6 showed a borderline association (p = 0.065), and R wave amplitude in V8 was not significant (p = 0.551). The model was globally significant (likelihood-ratio p < 0.0001) and showed high apparent discrimination (AUC = 0.944). Notably, four of the five variables selected in the primary H versus CaD analysis were also selected in the HCM-restricted analysis (R wave amplitude in lead II, aVL, and V6 and P wave amplitude in V8), with three of them retaining statistical significance in the final sensitivity model.
At enrollment, 29 of 57 cats in the CaD group were receiving cardiac medication, whereas 28 were untreated. In a sensitivity analysis restricted to CaD cats, none of the ECG variables retained in the final multivariable model differed significantly between treated and untreated cats. These findings did not provide evidence of a major association between treatment status and the principal ECG variables identified by the multivariable analysis.
P wave amplitude differed significantly in nearly all pairwise comparisons across all 13 leads in each group (Table 3, Table 4 and Table 5; Figure 3). Peripheral limb leads consistently showed higher P wave amplitudes compared to precordial leads. However, P wave amplitude in lead I did not usually differ from precordial leads in group H and NCD (p = 0.398–0.991), and it usually showed lower values compared to leads V5–V6 in the CaD group.
R/S ratios were assessed in precordial leads among the three groups (Appendix A, Table A4). In lead V1, the R/S ratio was predominantly <1 in all groups (57.5% in H group, 69.1% in CaD group, 62.5% in NCD group). In the left precordial leads (V2–V8), R/S was predominantly >1 in all groups.
4. Discussion
A 13-lead ECG system in cats has a moderate accuracy in differentiating between H cats and cats with CaD but not with NCD. Contrary to our initial hypothesis, however, absolute P wave amplitudes in peripheral limb leads were higher than in precordial leads, while among precordial leads, P wave amplitude in lead V1 had the lowest absolute value. Adding lead V8 likely does not improve the detection of sinus P waves, since no difference with other left precordial leads was observed in any of the three groups.
In humans, standard 12-lead ECG includes six unipolar precordial leads (V1–V6) distributed across the anterior and lateral walls of the chest [3]. V1 and V2 electrodes are placed on the fourth intercostal space at the right and left sternal border, respectively, V3 midway between V2 and V4, and V4 on the fifth intercostal space at the midclavicular line. V5 and V6 are positioned on the horizontal plane of V4, at the anterior axillary and midaxillary lines. Additional posterior leads (V7–V9) have been introduced to improve detection of electrical activity arising from regions of the left ventricle that are poorly represented by the conventional six precordial leads, in particular the inferobasal and posterolateral walls. Among them, V8 is placed over the tip of the scapula on the same horizontal plane as V6 [3]. In humans, V7–V9 increase the spatial coverage of the chest, particularly over the inferobasal and posterolateral left ventricular wall [21,22]. By analogy, the V8 position adopted in our protocol was intended to extend the coverage of the feline chest beyond the conventional six precordial leads and to optimize the recording angle for atrial signals projected toward the left dorsolateral thorax. Therefore, the V8 electrode was placed at the fifth left intercostal space at the level of the scapulohumeral joint.
Several publications have established reference values for feline peripheral limb leads and described the influence of body position on the recorded waveform [23,24,25]. A recent study [8] investigated the utility of selected precordial leads (V1, V2 and V4) in healthy cats and, similarly to our study, the authors observed lower P wave amplitudes in precordial leads, with the lowest values recorded in V1. In dogs, some precordial lead systems have also been used to better identify P waves. In one study, Santilli et al. [5] showed that relocating V1 to the right first intercostal space yielded more reliable detection of right atrial activity across different thoracic conformations. Kurogochi et al. [26] tested an extended precordial system including the posterior lead V10 and two bipolar leads (M6M1 and M6M2). On those leads, P wave amplitudes appeared higher compared to the conventional limb and precordial leads, similarly to the Lewis lead system employed in human patients [27]. In humans, the synthesized 18-lead ECG extends the standard system with the back leads V7–V9, and the P wave morphology in V7 has been shown to differentiate atrial arrhythmias arising from anatomically adjacent structures [28]. Based on our results, however, it appears that the electrical vector of sinus P waves in cats does not align with this direction; therefore, in cats, left precordial leads with or without cardiac disease do not improve the detection of these low-voltage events.
Among the ECG variables evaluated in the H group, QRS complex duration differed from reference values previously reported in the literature. Indeed, whereas an upper limit of normality of <40 ms is generally reported for healthy cats [2,16,19], the third quartile of QRS duration in lead II was 60 ms in our cohort of 40 healthy cats. Although our control population was not large, it was sufficiently sized to provide reliable descriptive values. Furthermore, all cats underwent echocardiographic examination to confirm the absence of structural cardiac disease, and all ECG measurements were performed by an experienced cardiologist who was blinded to the animals’ clinical status. One possible explanation for the longer QRS durations observed is the improved quality of contemporary digital ECG systems. Higher signal resolution, together with the recording settings used, may have facilitated the identification of low-amplitude deflections, particularly the Q and S waves, thereby resulting in slightly longer measured QRS durations compared with those reported in earlier studies. Nevertheless, the possibility that a small number of cats had subclinical intraventricular conduction disturbances cannot be completely excluded, and this may have contributed to a mild prolongation of QRS duration in our cohort. Future studies should consider that high-resolution ECG recordings in cats may yield longer measured QRS durations than those previously reported. This should be taken into account when interpreting QRS abnormalities, including wide QRS complex rhythms and bundle branch blocks.
Multiple ECG variables differed between the H and CaD groups, whereas none was able to discriminate between H and NCD cats. Of the 35 variables that differed between H and CaD on pairwise comparison, four ECG variables and age were retained as predictors of CaD in the multivariable model. These were R wave amplitude in lead II, R wave amplitude in lead aVL, R wave amplitude in V6, and P wave amplitude in V8. However, age and R wave amplitude in V6 were borderline significant and therefore likely less relevant; they were not retained as significant variables when a sub-analysis including only cats with the HCM phenotype was performed. In a recent study [29], R wave amplitude in lead II showed the highest discriminatory value between healthy cats and cats with compensated (pre-clinical) HCM, using a cut-off value of >0.26 mV. However, in the study of Gardner et al. [29], the accuracy of this variable was relatively low, since all cats in the diseased group were in the subclinical stage of disease. In our study, a 0.05 mV increase in R wave amplitude in lead II was associated with 1.54-fold higher odds of having CaD and 1.9-fold higher odds of having an HCM phenotype. In humans, increased R wave amplitude in aVL is recognized as a component of the Cornell voltage criteria for left ventricular hypertrophy [30] and has been associated with increased cardiovascular morbidity and mortality in the general population [31]. A similar interpretation may apply to feline structural heart disease, in which a thickened lateral left ventricular wall could shift voltage toward aVL. Although our study was not restricted to cats with left ventricular hypertrophy, HCM was the most prevalent type of cardiac abnormality, accounting for 33 of the 57 patients in the group, while 14 cats were equivocal for HCM. Indeed, maximal left ventricular wall thickness showed higher values in the CaD group as compared with the other two groups. Moreover, R wave amplitude in aVL remained significant also when only HCM cats were considered for the sub-analysis. Therefore, it is possible that increased R wave amplitude in aVL predicts left ventricular hypertrophy in cats as in humans. Similarly, R wave amplitude in precordial leads, V6 and V8, were borderline significant in predicting CaD and HCM. This supports the evidence that R wave amplitudes likely reflect the presence of left ventricular wall hypertrophy, as observed in other studies [29].
The P wave amplitude in V8 was identified as another predictor for CaD on multivariable analysis. We hypothesize that this reflects the high prevalence of left atrial enlargement in feline cardiomyopathy [14] and supports the contribution of left precordial leads to the assessment of atrial activity. This observation is in line with the findings of Santilli et al. [8], who reported higher P wave amplitudes in left than in right precordial leads in healthy cats. Since cats in the H group were younger than diseased cats, age was analyzed as a continuous variable in the multivariable analysis, and it resulted as a risk factor for CaD and HCM, along with the other already-mentioned ECG variables. This finding has been reported in previous studies, showing that older cats have a higher risk of developing HCM and highlighting the possible age-related increase in expression of an abnormal phenotype, as in people [32].
QT and QTc intervals did not differ significantly between the H and CaD groups in our population. This contrasts with the findings of Romito et al. [1], who reported prolongation of QT and QTc on a six-lead ECG in cats with left ventricular hypertrophy. Similarly, Bastos et al. [33] documented longer QT intervals and increased QT instability in cats with an HCM phenotype compared with healthy controls. Possible reasons for this discrepancy include the heterogeneity of our CaD group. Most cats in the CaD group had cardiomyopathy, with several phenotypes represented, while a smaller proportion presented with valvular or other cardiac diseases. In addition, cats were enrolled at variable disease stages. By contrast, Romito et al. [1] and Bastos et al. [33] specifically enrolled cats with a hypertrophic phenotype.
Our findings on the precordial R/S ratio were comparable to those described by Santilli et al. [8]. The R/S ratio was predominantly <1 in V1 across groups, while in left precordial leads (V3–V8) the R/S ratio was >1 in most cats. The predominant R/S patterns were detected among groups; the trend toward higher R/S ratios in V3–V8 in CaD is consistent with a leftward and dorsal shift of the QRS vector in the presence of left ventricular hypertrophy [19].
Several limitations should be acknowledged. First, the groups including diseased cats showed a variety of different cardiac and non-cardiac diseases. Moreover, some cats in the CaD group also suffered from non-cardiac and systemic diseases. However, we believe that this is more representative of the real-life conditions typical of daily practice. The intent was mainly to evaluate the discriminatory value of ECG between healthy and diseased cats and to assess if non-cardiac diseases might affect electrocardiographic variables. To overcome this limitation, we ran a sub-analysis comparing only H cats and cats from the CaD group that showed an HCM phenotype. The results were similar to those of the original analysis, retaining as significant predictors of disease almost all the same variables (namely R wave amplitude in leads II, aVL, V6, and V8 and P wave amplitude in V8).
Second, mild sedation was allowed for some cats, to reduce stress and improve the quality of the examination. Recent randomized studies in cats have shown that a single oral dose of gabapentin produces only minor changes in cardiovascular parameters [34,35] and does not significantly affect standard ECG variables [36]. Butorphanol was the only opioid sedative used in our study, while alfaxalone was not allowed. In one study of healthy cats, combined intramuscular administration of alfaxalone and butorphanol did not produce clinically relevant changes in echocardiographic measurements, but no effect on the ECG was assessed [37]. To our knowledge, no study has specifically evaluated the effect of butorphanol alone, or the combination of butorphanol with gabapentin, on ECG variables in cats. Our analysis showed that no association was present between sedation status and any ECG variable.
Some cats in the CaD group were receiving medications, such as diuretics and antiarrhythmics, that might affect some ECG variables, such as amplitude or heart rate. Although the principal ECG variables identified by the multivariable analysis did not differ between treated and untreated CaD cats in the sensitivity analysis, a treatment effect cannot be excluded. Treatment was not randomly assigned and was prescribed according to clinical indication; therefore, differences between treated and untreated cats may also reflect disease severity and other clinical characteristics. Consequently, the present study cannot fully disentangle disease-related ECG changes from treatment-related effects.
Third, although electrodes were placed by only two operators, following a standardized protocol, some small inaccuracies in limb or precordial lead position cannot be excluded. Anxious cats in which the forelimbs might not have been optimally extended, and overweight cats in which the intercostal spaces and sternum could be difficult to palpate, may have had incorrect electrode positioning. Additionally, to record 13 leads, the V1 electrode was replaced in the V8 position. Therefore, a simultaneous recording of all 13 leads was not possible. Slight changes in heart rate or stress conditions during the two phases of recording might have influenced the results.
Fourth, the population was recruited at a single referral cardiology service with one ECG machine and a specific setting and is therefore not necessarily representative of the general feline population. In particular, the sampling frequency of 500 Hz, while widely adopted across the literature, may have been insufficient to accurately resolve ECG waveforms in cats, whose characteristically elevated heart rates demand greater temporal resolution.
Fifth, this study included only cats in sinus rhythm. It was not our aim to assess the accuracy of ECG in diagnosis any specific arrhythmia nor the adjunct value of precordial leads in identifying non-sinus P waves. It is indeed possible that in some cases, when P’ waves are not visible in peripheral limb leads, precordial leads might help in visualizing these low voltage events, especially if the P’ wave axis is significantly deviated, like in the case of atrial tachycardia or retrograde P’ waves during bypass mediated or junctional rhythms.
Finally, regarding the multivariable model, the apparent discrimination of the model derived from the original dataset was high (AUC = 0.884). However, because predictor selection and model estimation were performed in the same cohort, this value is expected to overestimate the performance. Internal bootstrap validation was therefore performed by repeating the entire variable-selection procedure in each of 2000 bootstrap samples. After correction for optimism, the AUC decreased to 0.795. Thus, although part of the apparent performance was attributable to model development in the same dataset, the modelling procedure retained acceptable-to-good discriminative ability after internal validation. Bootstrap selection frequencies also showed that the composition of the model was not completely stable. R wave amplitude in lead II showed the greatest stability and was selected in approximately two-thirds of bootstrap samples, whereas P wave amplitude in V8 was retained in approximately half. The remaining variables were selected less consistently, particularly R wave amplitude in V6. This variability probably reflects the relatively limited sample size and the presence of several ECG measurements providing partially overlapping information. Consequently, the individual variables retained in the final model should not be regarded as definitive predictors of cardiac disease but rather as variables associated with cardiac disease in this cohort whose relevance requires confirmation in independent populations. The sensitivity analysis restricted to cats with a definite HCM phenotype provided additional support for the primary multivariable findings. Despite restricting the analysis to a smaller but clinically more homogeneous cardiac population, four of the five variables selected in the primary analysis were retained by the forward-selection procedure. In particular, R wave amplitude in lead II, aVL, and V6 and P wave amplitude in V8 remained significantly associated with disease status in the final HCM-restricted model, with R wave amplitude in V8 as an additional significant variable. This consistency suggests that the associations observed for these ECG measurements were not solely driven by the heterogeneity of cardiac abnormality included in the overall CaD group.
5. Conclusions
In conclusion, a 13-lead ECG system including peripheral limb leads and seven precordial leads can be applied in cats. Peripheral limb leads have the highest P wave amplitudes among all leads. V8, however, provides higher P wave amplitudes than V1 and may assist in the assessment of atrial electrical activity. The present study also provides descriptive values from healthy adult cats for peripheral limb leads and, as a novel contribution, for seven precordial leads. Some precordial and peripheral limb lead variables, in particular R wave amplitude in lead II and aVL and P wave amplitude in V8, were predictors of CaD, while no ECG parameter differed between cats with NCD and H cats. Moreover, the same variables were useful in differentiating between H cats and cats with an HCM phenotype.