兽研猫传染性腹膜炎（FIP）的生前临床和临床病理学诊断仍然具有挑战性，尤其是当积液偏离典型的富含蛋白质、低至中等细胞度的表型时。FIP积液指数定义为总核细胞数差值（ΔTNC）与白蛋白与球蛋白比值（A/G）的比值，最初在50份积液中开发并评估。随后欧洲猫病咨询委员会（ABCD）指南建议在更大队列和确诊FIP的猫中进行评估。本研究回顾性评估了预设阈值（≥7.54）在176份独立的猫积液样本中的应用，这些样本在Sysmex XN-1000V上分析，并通过积液FCoV RT-qPCR进行分类，未独立确认FIP状态。相对于该分子参考，FIP积液指数的敏感性为97.8%（95%置信区间[CI]，92.4–99.7%），特异性为88.1%（95% CI，79.2–94.1%），受试者工作特征曲线下面积（AUC）为0.962（95% CI，0.928–0.995）。其判别能力超过单独ΔTNC（AUC，0.947；ΔAUC，0.014；p = 0.012）和单独A/G（AUC，0.887；ΔAUC，0.074；p = 0.012）。在渗出液中，敏感性保持100%（95% CI，80.5–100%），特异性为82.8%（95% CI，64.2–94.2%）。在参考阴性的积液中，假阳性分类在脓毒性样本中比非脓毒性样本中更常见（6/19 vs. 4/65；比值比，7.04；95% CI，1.74–28.53；p < 0.01）。这些发现表明，预设的FIP积液指数阈值在更广泛的猫积液谱中对分子参考阳性保持了非常高的敏感性和强判别能力。未来的研究应在序贯体内诊断算法中评估FIP积液指数，利用阴性结果支持排除诊断，阳性结果提示通过免疫细胞化学和/或免疫组织化学检测FCoV抗原来确认。
1. 引言
猫传染性腹膜炎（FIP）是一种系统性炎症性疾病，在少数感染猫冠状病毒（FCoV）的猫中发生，仍然是猫医学中最具挑战性的诊断之一[1,2,3]。有效抗病毒治疗的可及性进一步增加了建立早期可靠诊断以允许及时治疗的临床重要性[4,5,6,7,8,9]。当存在积液时，应优先进行积液分析，因为细胞学、生化评估和直接检测FCoV RNA或抗原的组合提供的诊断信息远多于外周血检测[1,10,11,12]。特征性组织病理学病变连同通过免疫组织化学（IHC）检测病变内FCoV抗原构成确定性FIP诊断的参考标准，尽管组织采样具有侵入性，并且在临床患病猫中可能不切实际[1,2,10,11,12]。积液样本的FCoV逆转录定量聚合酶链反应（RT-qPCR）可快速获得，并已显示出高敏感性和特异性，在多项研究中通常达到100%[2,13,14,15,16,17,18,19,20,21]。Longstaff等[20]报告积液中敏感性为85%、特异性为100%，而Doenges等[21]报告无细胞体腔积液中敏感性为88.9%、特异性为100%，相对于包括IHC检测FCoV抗原的疾病分类。尽管组织病理学结合病变内FCoV抗原检测仍是参考标准，但本研究选择积液RT-qPCR作为分子参考分类，因为积液采样比组织活检侵入性更小，结果可快速获得，且该检测适用于常规诊断工作流程。
传统上，FIP相关积液通常富含蛋白质，呈黄色至稻草色，黏稠且有粘性，常伴有纤维蛋白或凝块形成，并常呈现改性漏出液表型[3,22,23,24]。在细胞学上，它们通常含有非退行性中性粒细胞、巨噬细胞或单核细胞和少量淋巴细胞的混合性脓肉芽肿性细胞群，背景为蛋白质性颗粒状[3,22,25,26,27]。然而，生物学谱比这种传统描述所暗示的更广泛。当前的FIP诊断指南认识到，FIP中总核细胞计数（TNCC）可显著升高，包括超过20 × 109/L的值，并且继发性细菌性腹膜炎可与FIP共存[1,2,3,28,29]。从诊断角度来看，这意味着积液表型应谨慎解读：渗出性模式不能排除FIP，而与FIP无关的严重炎症性疾病可能具有一些通常被认为支持该疾病的生化特征[1,22,25,28]。
在廉价的基于积液的诊断检测中，白蛋白与球蛋白（A/G）比值和Rivalta试验仍被广泛使用[2,3,30]。然而，两者均具有重要局限性。低A/G比值反映FIP的炎症性蛋白特征，但不具有疾病特异性；而Rivalta反应阳性可见于脓毒性腹膜炎和淋巴瘤相关积液[2,17,31,32,33]。诊断指南将低积液A/G比值归入“较可能”而非“确诊”类别，其中<0.4增加怀疑，>0.8使FIP可能性降低[2,3,30,32]。Sysmex XN-1000V的白细胞分类（WDF）和白细胞核（WNR）分析通道所获得的细胞性测量结果之间的不一致提供了进一步的诊断信号[1,11]。称为总核细胞差值（ΔTNC）的参数在Sysmex血液学平台上分析的猫积液中显示出高诊断准确性[34,35]。
FIP积液指数由我们的研究组于2024年提出，旨在将这种细胞通道现象与A/G比值所包含的生化信息相结合[1]。在50份积液的推导队列中，FIP积液指数阈值≥7.542相对于积液FCoV RT-qPCR状态产生了96.3%的敏感性和100%的特异性[1]。重要的是，该初步研究排除了脓毒性、不明确和不结论性的细胞学样本，从而评估了相对受限的诊断谱。随后详细的2026年欧洲猫病咨询委员会（ABCD）FIP指南强调FIP积液指数具有前景，特别指出需要在更大数量的样本和更明确确诊的FIP病例中进行评估，之后才能常规推荐[30]。
验证诊断指数的主要挑战不仅是在类似于推导队列的人群中重现其性能，而是确定预设阈值在面对常规实践中遇到的异质性炎症和肿瘤性疾病时是否仍具有信息价值。在每个新数据集中重新优化阈值会夸大表观性能，不构成严格的验证。因此，本研究保留≥7.54作为主要的预设临床阈值，并将任何数据驱动的阈值分析视为探索性分析。
主要目的是在更大规模、扩展诊断谱的猫积液队列中，以积液FCoV RT-qPCR状态作为分子参考分类，评估FIP积液指数的诊断性能。次要目的为：（i）比较FIP积液指数与其各单独组分ΔTNC和A/G的连续判别能力；（ii）比较预设FIP积液指数阈值与先前发表的ΔTNC、A/G和Rivalta试验阈值；（iii）确定在渗出性积液中是否保留敏感性，渗出性积液超出经典的改性漏出液表型，可能呈现更高的误分类风险；以及（iv）研究脓毒性炎症作为假阳性结果的来源。我们假设FIP积液指数在积液表型中保留高敏感性，而严重脓毒性炎症将构成特异性的生物学上合理的局限性。
2. 材料与方法
2.1. 研究设计、场所和诊断记录
本回顾性、扩展诊断谱的诊断准确性研究使用2025年3月1日至2026年3月1日期间提交至CEDIVET兽医实验室（葡萄牙波尔图）进行常规诊断检查的猫积液样本进行。符合条件的样本包括不同细胞学和生化表型的积液，包括改性漏出液、渗出液和脓毒性炎症性积液。所有性能估计均相对于积液FCoV RT-qPCR分类计算，而非相对于独立确认的FIP状态。
当有有效的FCoV RT-qPCR结果以及计算FIP积液指数所需的所有实验室测量值时，积液记录符合纳入条件，即来自WDF（TNCC-WDF）和WNR（TNCC-WNR）通道的总核细胞计数以及积液A/G比值。TNCC-WDF和TNCC-WNR测量分别使用Sysmex XN-1000V分析通道及Lysercell WDF（货号CV377552）、Fluorocell WDF（货号AA325279）、Lysercell WNR（货号DWX-50102）和Fluorocell WNR（货号M375584）试剂获得（Sysmex Corporation，日本神户）。相应的比较和亚组分析还需要Rivalta试验结果、积液表型和脓毒性分类。
样本采集后置于乙二胺四乙酸（EDTA；4 mL；货号454209，Greiner Bio-One，Kremsmünster，奥地利）管和血清管（4 mL；货号454092，Greiner Bio-One，Kremsmünster，奥地利）中，并在采集后18–24小时内处理，遵循原研究使用的分析方案[1]。当分子结果分析无效、可用样本量不足以完成所需分析，或推导主要指标检测（即TNCC-WDF、TNCC-WNR或积液A/G比值）所需的任何测量值缺失时，记录被排除。具体而言，扩增对照无效、样本量不足或TNCC-WDF、TNCC-WNR或A/G比值缺失的样本被排除。RT-qPCR结果按照制造商的定性标准判读：扩增对照有效且分析运行内目标扩增的样本归类为分子参考阳性，而对照有效但无目标扩增的样本归类为分子参考阴性。本研究未单独定义临界或不确定的分子类别，分析无效的结果被排除在诊断准确性分析之外。没有符合初始入选标准的记录因上述原因被排除。与原推导研究不同，本研究有意保留脓性积液，以在常规诊断实践中遇到的更广泛的炎症条件下评估FIP积液指数。
统计单位为单个积液样本。全部176个样本代表来自176只个体猫的独立观察，没有猫在主要分析中贡献超过一个样本。在预定义研究期内符合纳入标准的所有合格样本均被纳入，主要诊断分析所用变量未进行插补。本研究按照诊断准确性报告标准（STARD）[36]进行报告。
2.2. 实验室分析与积液分类
根据情况对直接涂片、压片制备或细胞离心标本进行细胞学检查，使用RAL染色机（CellaVision RAL Diagnostics，Martillac，法国）以罗曼诺夫斯基型染色液染色[1]。
积液根据总蛋白浓度、细胞量和细胞学特征进行分类。传统框架将体液分为漏出液、改良漏出液或渗出液[1,37]。先前建立的标准包括：漏出液为总蛋白< 2.5 g/dL且TNCC < 1500个细胞/μL；改良漏出液为总蛋白2.5–7.5 g/dL且TNCC 1000–7000个细胞/μL；渗出液为总蛋白> 3.0 g/dL且TNCC > 7000个细胞/μL[1]。
总蛋白和白蛋白浓度分别使用Beckman Coulter AU680e分析仪（Beckman Coulter，Cassina de' Pecchi，意大利）通过双缩脲法（货号OSR6132，Beckman Coulter®，Brea，CA，美国）和溴甲酚绿法（货号OSR6102，Beckman Coulter®，Brea，CA，美国）测定。球蛋白浓度按总蛋白减去白蛋白计算，A/G比值按白蛋白除以球蛋白计算。Rivalta试验按照先前描述的方案进行和判读[1,31,33]。
Nucleated-cell measurements were obtained using the Sysmex XN-1000V haematology analyser (Sysmex Corporation, Kobe, Japan) in whole-blood mode. ΔTNC was operationally defined as the ratio (1):
[内容请查看原文]
采样积液的原解剖部位记录为腹腔、胸腔或心包腔。这种样本水平的分类并不意味着每只猫的积液局限于单一腔隙。炎症性积液内细胞内细菌的细胞学证据支持分类为脓性，微生物培养在可获得时被视为佐证。
对于预先设定的表型分析，将改良漏出液与属于渗出液家族的积液进行比较。较少见的表型，包括乳糜性、漏出性和出血性积液，仍纳入总体分析。
2.3. 分子分析与参考分类
使用Promega AX9760（Promega Corporation，Madison，WI，USA）全血定制核酸提取方案从积液样本中提取病毒RNA，该方案针对体液样本进行了调整，并在KingFisher™ Flex系统（Thermo Fisher Scientific，Waltham，MA，USA）上实施，遵循先前描述的实验室流程[1]。使用NZYtech猫冠状病毒RT-qPCR试剂盒（Cat. MD05601，NZYtech®，Lisbon，Portugal）在QuantStudio™ 5实时PCR系统（Thermo Fisher Scientific，Waltham，MA，USA）上检测FCoV RNA，按照制造商说明进行操作。热循环包括50 °C逆转录20 min、95 °C聚合酶激活2 min，以及40个循环的95 °C变性5 s和60 °C退火/延伸30 s [1]。每个反应均包含内部提取对照，每次分析运行均包含适当的阳性和阴性对照。仅当内部提取对照适当扩增、阳性对照显示扩增且阴性对照未观察到靶标扩增时，分子结果才被认为分析有效。未引入研究特定的定量循环截断值。有效样本根据制造商的定性标准进行分类，定量循环（Cq）< 40且具有适当S形曲线的靶标扩增判定为阳性。无靶标扩增的有效样本判定为阴性。无效结果被排除在诊断准确性分析之外。定性阳性标准不区分低病毒载量与高病毒载量。符合相同有效性和阳性标准的晚期扩增未赋予单独的研究特定类别。
积液FCoV RT-qPCR定性结果构成分子参考分类。分组分配仅基于FCoV RT-qPCR，就FIP状态而言为推定性的。积液中FCoV RNA的检测并非确定性FIP的无误差参考标准；因此诊断性能估计描述的是相对于分子状态的区分和分类能力，而非绝对疾病状态[38,39,40]。尽管组织病理学结合病灶内FCoV抗原检测仍为参考标准，但本研究选择积液RT-qPCR作为分子参考分类，因为积液采样比组织活检创伤更小、结果可快速获得且该检测适用于常规诊断流程。此选择并不确立独立的疾病确认。该检测可检测FCoV但不能区分猫肠道冠状病毒（FECV）与FIP相关变异株（FIPV），且单独的积液RT-qPCR阴性结果不能排除FIP [30]。
2.4. FIP积液指数与预设阈值
The FIP Effusion Index was calculated according to Equation (2) originally proposed by Lopes et al. [1]:
[内容请查看原文]
FIP积液指数在应用任何诊断阈值之前作为连续变量计算。主要指数检测在预设临床阈值≥7.54下评估，该值为先前研究中独立于本研究数据集得出的≥7.542阈值的取整形式，随后被2026年ABCD指南引用[1,30]。历史未取整阈值≥7.542作为敏感性分析单独评估，未用于重新定义主要操作点。
先前发表的比较阈值分别为ΔTNC ≥ 4.903、ΔTNC ≥ 2.081和A/G ≤ 0.56 [1]。Rivalta试验阳性作为二分类比较指标进行评估。
主要诊断结局为敏感性和特异性。次要指标包括总体准确性、阳性和阴性似然比（LR+和LR−）以及诊断比值比（DOR）。在本研究中，敏感性、特异性、准确性、似然比以及真/假阳性或阴性分类均相对于积液FCoV RT-qPCR定义，而非独立确立的FIP状态。
2.5. 统计分析
连续变量以中位数和四分位距（IQR）汇总，分类变量以计数和百分比汇总。FCoV RT-qPCR阳性和阴性积液之间的差异使用双侧Mann–Whitney U检验评估，秩双列相关性作为效应量指标报告。
预设阈值下的诊断性能通过二乘二列联表评估。敏感性和特异性及准确性以精确95%置信区间（CIs）计算，同时计算LR+、LR−和DOR。FIP积液指数、ΔTNC和A/G比的连续区分能力使用受试者工作特征（ROC）曲线和曲线下面积（AUC）评估。A/G因较低值支持FIP而进行方向反转。AUC使用配对DeLong检验进行比较[41]，对三个成对比较应用Holm校正。
固定阈值下敏感性和特异性的差异使用精确McNemar检验评估，并应用Holm校正。预设亚组分析比较改良漏出液与渗出液以及脓毒性与非脓毒性积液；腹部和胸腔积液进行探索性分析。独立比例使用Fisher精确检验进行比较。
Youden指数仅用于阈值一致性的探索性评估。AUC和探索性阈值稳定性通过3000次保留参考类别比例的分层bootstrap重采样进行评估。未进行事后功效计算。所有检验均为双侧检验，p < 0.05被认为具有统计学显著性[42]。
统计分析使用JMP® 18.0.0版（SAS Institute Inc.，Cary，NC，USA）、DATAtab®（numiqo e.U.，Graz，Austria，2026）、MedCalc® Statistical Software 20.006版（MedCalc Software Ltd.，Ostend，Belgium）和R 4.6.1（R Foundation for Statistical Computing，Vienna，Austria）进行。
3. 结果
3.1. 研究人群
分析数据集包含176条积液记录，其中92条（52.3%）为FCoV RT-qPCR阳性，84条（47.7%）为阴性。改良漏出液为主要表型（117/176；66.5%），其次为渗出液（46/176；26.1%）；13条记录（7.4%）包括7例乳糜性、5例漏出性和1例出血性积液。23/176（13.1%）次观察记录到脓毒性炎症。采样的积液包括118例腹腔积液（67.0%）、55例胸腔积液（31.3%）和3例心包积液（1.7%）。所有记录均有解剖位置和性别信息。信号特征和主要液体特征总结于表1。
3.2. 按参考分类的连续性积液标志物
分子参考阳性的积液较参考阴性积液具有更高的总蛋白和球蛋白浓度以及更低的A/G比值（表2）。最大的单变量效应量见于FIP积液指数（秩双列r = 0.923）和ΔTNC（r = 0.894）。参考阳性积液中FIP积液指数的中位值为45.90（IQR 25.41–94.81），参考阴性积液中为1.21（0.86–2.14）（p < 0.001；秩双列r = 0.923）。相应的ΔTNC中位值分别为19.09（9.82–37.00）和0.98（0.87–1.17）（均p < 0.001）。球蛋白（r = 0.779）和A/G（r = −0.775）也显示出显著分离。相比之下，单独TNCC-WDF提供的分离很小（中位值1.987 vs. 2.800 × 109/L；p = 0.090；r = −0.148）。图1显示了锁定阈值下的FIP积液指数分布。
3.3. 预设FIP积液指数阈值的初步验证
应用独立建立的阈值≥7.54，未在本数据集中进行优化，正确分类了90/92例参考阳性和74/84例参考阴性积液（TP/FP/FN/TN，90/10/2/74）。敏感性为97.8%（95% CI，92.4–99.7%），特异性为88.1%（95% CI，79.2–94.1%），总体准确率为93.2%（95% CI，88.4–96.4%）。LR+为8.22（95% CI，4.59–14.71），LR−为0.025（95% CI，0.006–0.097），对应的DOR为333（95% CI，70.7–1567.4）。
使用精确的历史阈值≥7.542而非其四舍五入的临床形式导致额外一例假阴性分类：敏感性为96.7%，而特异性保持88.1%。因此，所有主要和亚组分析均保留临床阈值≥7.54。
3.4. 连续性判别和增量价值
连续性FIP积液指数显示出优异的判别能力，AUC为0.962（95% CI，0.928–0.995）。ΔTNC和A/G的AUC分别为0.947（95% CI，0.910–0.984）和0.887（95% CI，0.837–0.937）（图2）。由于ROC AUC评估连续性标志物，ΔTNC的AUC独立于固定截断值。ΔTNC ≥ 4.903和≥2.081时的操作特征分别报告于表3。固定FIP积液指数阈值下ΔTNC与A/G之间的互补二维关系见图3。在配对DeLong分析中，FIP积液指数超过ΔTNC 0.014个AUC单位（95% CI，0.004–0.025；p < 0.01；Holm校正p = 0.012），超过A/G 0.074（95% CI，0.024–0.125；p < 0.01；校正p = 0.012）。ΔTNC也超过A/G（ΔAUC，0.060；95% CI，0.004–0.115；p = 0.034；校正p = 0.034）。差异由未四舍五入的AUC估计值计算得出。因此，将ΔTNC与A/G联合较单独ΔTNC产生了统计学上可证明但数量上适度的改善。
3.5. 配对固定阈值比较
在固定截断值下，FIP积液指数正确识别了6例被ΔTNC ≥ 4.903遗漏的参考阳性积液，而没有参考阳性积液被ΔTNC正确识别但被该指数遗漏。精确McNemar p值为0.031。然而，在预设固定阈值比较族进行Holm校正后，该差异未保持显著性（校正p = 0.125）。相比之下，FIP积液指数相对于A/G ≤ 0.56的敏感性优势保持显著（校正p < 0.01）。相对于Rivalta试验的主要固定阈值优势为特异性：在参考阴性记录中，17例被FIP积液指数正确分类但Rivalta试验为阳性，而无相反方向的不一致（精确p < 0.001；校正p < 0.001）。
3.6. 改良漏出液之外的性能
预设阈值在两种主要积液表型中均保持了高敏感性。在117例改良性漏出液中，敏感性为97.3%（95% CI，90.7–99.7%），特异性为92.9%（95% CI，80.5–98.5%），AUC为0.970。在46例渗出液中，全部17例参考阳性记录均为FIP积液指数阳性，观察到的敏感性为100%（95% CI，80.5–100%）；特异性为82.8%（95% CI，64.2–94.2%），AUC为0.947（95% CI，0.876–1.000）。改良性漏出液与渗出液之间的敏感性无差异（Fisher p = 1.000）。改良性漏出液的特异性比渗出液高10.1个百分点（差值，+10.1个百分点；95% CI，−5.2–28.0；Fisher p = 0.258）。这些结果未提供证据表明渗出液分类会导致敏感性丧失。在渗出液样本中，26例非脓毒性积液（14例参考阳性、12例参考阴性）均未发生误分类，而17例参考阴性脓毒性渗出液中有5例为FIP积液指数阳性。尽管如此，脓毒性渗出液亚组的AUC仍保持在0.941，提示尽管一部分严重炎症性积液超过了固定阈值，该连续标志物仍具有很强的信息价值。鉴于亚组样本量小且部分分层存在零单元格，这些估计值应通过其置信区间来解读，而不应视为亚组分类完美的证据。
3.7. 脓毒性炎症与假阳性分类
脓毒性炎症是与特异性降低相关的主要可识别背景。在所有分子参考阴性积液中，假阳性FIP积液指数结果在脓毒性记录中发生6/19例（31.6%），而在非脓毒性记录中发生4/65例（6.2%）。因此，脓毒性积液的特异性为68.4%，非脓毒性积液为93.8%，对应差值为−25.4个百分点（95% CI，−48.1至−7.1；Fisher p < 0.01）。脓毒性参考阴性积液发生假阳性分类的几率约为非脓毒性参考阴性积液的七倍（OR，7.04；95% CI，1.74–28.53）。按分子参考和脓毒性状态分组的观察值分布见图3。
值得注意的是，脓毒性炎症并非必然导致FIP积液指数升高（表4）。图4和图5展示了一例具有代表性的FCoV RT-qPCR阴性脓毒性渗出液，其被正确分类为低于预设阈值。尽管细胞学检查显示明显的炎症性改变，该积液的ΔTNC为1.00，A/G比为1.19，所得FIP积液指数为0.84。该示例说明，即使存在明显的脓毒性化脓性炎症，也不一定会重现与高FIP积液指数相关的通道间不一致。
3.8. 探索性阈值稳定性与Bootstrap分析
在完整验证队列中进行的探索性Youden分析显示，预设阈值7.54达到了最大Youden J值（0.859），敏感性为97.8%，特异性为88.1%。算法推导的最佳阈值为6.300，其敏感性、特异性和Youden J值与预设阈值相同。在3000次分层Bootstrap重采样中，算法推导最佳阈值的中位数为6.300（IQR，6.171–6.362），第2.5至第97.5百分位范围较宽，为3.62–17.42，表明数据驱动的截断值选择存在明显的有限样本不稳定性。Bootstrap AUC分布更为稳定（中位数，0.964；百分位范围，0.924–0.990）。这些探索性发现表明，预设阈值在验证队列中达到了最大Youden性能，而Bootstrap结果则说明选择单一数据驱动最佳阈值的不稳定性；这些结果并不回溯性验证或重新推导预设截断值（图6）。
4. 讨论
本研究的目的并非重新定义截断值，而是确定先前建立的FIP积液指数及其预设阈值[1]在一个时间上不同、规模大得多且涵盖更广泛积液谱的队列中是否仍具有信息价值。研究得出了四项具有临床意义的发现。第一，在规模超过原始推导系列[1]三倍以上的队列中，≥7.54的阈值保持了极高的敏感性（97.8%；95% CI，92.4–99.7%）、较强的总体判别能力（AUC，0.962；95% CI，0.928–0.995）和333的诊断比值比（DOR）（95% CI，70.7–1567）。第二，复合FIP积液指数在区分参考阳性与参考阴性积液方面优于单独使用ΔTNC或A/G，尽管相对于ΔTNC的增益在数量上较为有限。第三，渗出液中保留了高敏感性，表明渗出液表型并未消除该指数所捕获的分析信号。第四，脓毒性炎症是与特异性降低相关的最明显背景。这些观察结果既界定了FIP积液指数的潜在临床价值，也界定了其临床上可识别的局限性，但并未将分子参考阳性等同于独立确诊的FIP。
验证策略是本研究的重大优势。FIP积液指数公式和临床阈值均在独立的推导队列中建立，并在分析本研究数据之前锁定[1]。这一区别很重要，因为事后选择阈值可能产生过于乐观且无法推广的估计[43,44]。因此，时间上的分离、在预定义期间纳入所有符合条件的猫以及有意保留炎症性竞争诊断，构成了对可迁移性更严格的检验。探索性Youden分析显示，锁定的阈值7.54（随后在ABCD指南中再现[30]）达到了与算法推导的最佳阈值6.300相同的最大Youden J值，灵敏度和特异度完全相同，从而支持与先前建立的阈值的一致性，而非对其回顾性重新推导。Bootstrap算法推导的最佳阈值中位数为6.300，百分位范围较宽（3.62–17.42），进一步说明在有限验证样本中数据驱动截断值选择的不稳定性，以及为何已建立的临床阈值不应被反复重新校准。
本研究结果扩展而非简单重现了原始FIP积液指数研究[1]和早期ΔTNC文献[10,34,35]。在50例积液的推导队列中，≥7.542阈值产生96.3%的灵敏度和100%的特异度[1]。在本研究中，灵敏度相似，但特异度较低，为88.1%（95% CI，79.2–94.1%）。这些点估计不应被解释为可直接互换，因为本样本量更大，且纳入了在推导期间被限制或排除的脓性、渗出性及其他诊断困难的积液。早期XT-2000iV研究已确立ΔTNC为高度鉴别性的积液标志物[10,34,35]，而随后XT-2000iV与XN-1000V的比较显示，尽管诊断分类大体相似，但存在系统性的分析仪间差异[11]。因此，当前结果在规定的分析前和分析条件下，特别加强了FIP积液指数在XN-1000V上的证据。因此，不应假定ΔTNC值或FIP积液指数阈值可转移到另一台分析仪而不进行平台特异性验证。
配对分析有助于解释为何整合是有用的。ΔTNC在操作上为TNCC-WDF与TNCC-WNR的比值，捕捉由不同试剂和测量环境引起的通道间不一致，而A/G描述白蛋白相对于球蛋白在炎症蛋白环境中的分布。富含蛋白的FIP积液、单核/巨噬细胞驱动的炎症和血管通透性改变已有充分文献记载[45,46,47]，而纤维蛋白或蛋白相关的细胞聚集或在一个分析通道中的截留已被提出是ΔTNC的促成因素[10,11,34]。然而，XN-1000V上的确切物理化学机制尚未通过实验解决，这一解释应保持为假设，而非作为已确立的分析仪行为呈现。通过结合部分互补的细胞/分析信息和生化信息，FIP积液指数较ΔTNC的AUC增加0.014（95% CI，0.004–0.025；Holm校正p = 0.012），较A/G增加0.074（95% CI，0.024–0.125；校正p = 0.012）。较ΔTNC的增益有统计学支持但幅度不大。在固定阈值下，FIP积液指数检出了6例被ΔTNC ≥ 4.903漏诊的参考阳性积液，但该灵敏度比较在多重性校正后未保持显著（校正p = 0.125）。整合不同积液测量值的比值可能改善诊断鉴别这一更广泛的原则，也得到先前评估积液乳酸脱氢酶（LDH）/TNCC比值研究的支持，尽管该标志物与当前FIP积液指数在分析或临床上不可互换[48]。
渗出液中的诊断性能是本验证的核心进展。渗出性FIP传统上与富含蛋白、低至中等细胞数的改良漏出液相关，但当前指南和猫积液系列均显示细胞学类别与病因之间存在大量重叠[12,30,49,50,51]。传统漏出液、改良漏出液和渗出液标签描述的是液体特征；它们既非诊断性也非疾病特异性。全部17例参考阳性渗出液均为FIP积液指数阳性，对应观察灵敏度为100%，但95% CI为80.5–100%且分母较小，无法声称完美性能。渗出液中的特异度为82.8%（95% CI，64.2–94.2%），数据未提供证据表明灵敏度与改良漏出液不同。因此，仅凭渗出液分类既不应将FIP排除在鉴别诊断之外，也不应使指数的解读失效。反之，渗出性FIP积液指数阳性的积液不应被推定为代表FIP，尤其是当存在细菌性炎症的细胞学证据时。
因此，与推导队列[1]相比总体特异性较低具有科学参考价值。即使检测方法本身未变，诊断性能也会随病例构成、竞争性疾病及疾病严重程度分布而变化[44,52]。在本队列中，非脓毒性积液中特异性为93.8%（95% CI，85.0–98.3%），而脓毒性积液中为68.4%（95% CI，43.5–87.4%）。因此，较广泛的队列比受限的推导集是更难且更具临床相关性的挑战，但不应假定其代表所有预期使用人群。结果将观察到的大部分特异性损失定位在明确的炎症背景下，而非支持FIP积液指数普遍丧失判别力。
在84份参考阴性积液中，假阳性分类发生于6/19（31.6%）脓毒性样本和4/65（6.2%）非脓毒性样本，绝对特异性差异为−25.4个百分点（95% CI，−48.1至−7.1），未调整比值比为7.04（95% CI，1.74–28.53）。这确立的是关联，而非因果关系或独立于积液表型。该分析为单变量分析，且脓毒状态集中于渗出液。球蛋白富集、低A/G、纤维蛋白含量及细胞-试剂行为方面的重叠提供了合理解释，但这些机制未被直接测量[40,45,46,53,54]。同样，脓毒症并非总是产生高FIP积液指数：代表性脓毒性渗出液具有明显的化脓性炎症，但ΔTNC为1.00、A/G为1.19、FIP积液指数为0.84。因此，仅严重的细胞学炎症不足以重现完整特征。临床上，FIP积液指数阳性且含退变中性粒细胞和胞内细菌的积液需要更加谨慎，并结合细胞学、可用的细菌培养、直接FCoV检测及完整临床背景进行综合判断[30,55]。不应简单地将脓毒性积液排除在解读之外，也不应自动将细菌性炎症视为与FIP互斥，因为可发生继发性细菌并发症[30]。
似然比比单独的灵敏度和特异性更能有效地转化这些操作特征[56]。在本队列中，LR−为0.025（95% CI，0.006–0.097）尤为显著。假设积液RT-qPCR阳性的验前概率为50%，则验前比数为1.0。乘以0.025得到验后比数0.025，验后概率约为2.4%。因此，在该假设下，FIP积液指数阴性结果将显著降低分子参考阳性的概率。相比之下，在假设验前概率为10%时，LR+为8.22，将比数从0.111变为0.913，相当于验后概率约为47.7%：显著增加，但并非确定性结果。这些示例涉及的是此处使用的分子参考，而非组织病理学确诊的FIP。
FIP积液指数还应与常规检测并列而非对立定位。Rivalta试验和A/G仍廉价、快速且广泛可及，且均不需要指定的血液学平台[2,3,30,55]。在现有配对数据中，Rivalta试验保持高灵敏度（96.7%）但特异性较低（67.9%）。FIP积液指数正确分类了17份Rivalta阳性的参考阴性积液，且无相反方向的不一致。低A/G在确诊FIP中常见，但也伴随其他炎症状态，限制了其单独特异性[17,45,55]。在TNCC-WDF、TNCC-WNR和积液生化已可获得的情况下，其整合因此比任一单独组分提供更大的判别力，并可能较Rivalta试验提高特异性。积液巨噬细胞上的直接FCoV抗原检测提供不同信息，但研究间也显示出可变操作特征[29,57,58,59]。最近一项多重荧光免疫细胞化学诊断临床试验同样得出结论，该检测为补充性而非确定性[59]。总体而言，这些数据支持多模式诊断工作流程，而非以任何单一积液检测替代临床推理。
参考分类的选择为当前推论设定了重要边界。积液FCoV RT-qPCR具有临床价值，因为病毒RNA在FIP相关积液中通常丰富[60]。既往研究报告的敏感性分别为85.0%和88.9%，特异性相对于各自疾病分类为100%[20,21]。然而，RT-qPCR检测的是FCoV RNA而非定义性组织病变，并非完美的参考标准[20,21,24,30,55]。积液巨噬细胞中病毒抗原的免疫细胞化学和免疫荧光检测同样不完美[29,57,58,59]。2026年ABCD指南指出了FIP积液指数的两个未解决的证据需求：在更大样本中评估以及在确诊FIP的猫中评估[30]。本研究通过在176只猫中进行时间验证、涵盖更广泛的积液表型、纳入脓毒性和渗出性竞争病因以及应用原始锁定阈值，在很大程度上解决了第一个问题。但并未完全解决第二个问题，因为分类并非基于独立判定的疾病诊断。因此，下一步应超越进一步的分析验证，转向在序贯多变量诊断框架内对FIP积液指数进行前瞻性多中心疾病层面评估，其中其高敏感性支持初始排除FIP，而阳性病例则接受更具特异性的确认检测，特别是通过积液来源细胞块免疫细胞化学和/或组织病变免疫组织化学检测FCoV抗原。这种方法可在筛查阶段最大化敏感性，在确认阶段最大化特异性，从而有望提高体内FIP诊断的可靠性。
4.1. 局限性
若干局限性需要明确考虑。最重要的是，参考分类为积液FCoV RT-qPCR而非组织病理学结合病变内FCoV抗原证明。尽管积液RT-qPCR在适当的临床背景下信息价值很高，但它并非确定性FIP诊断的完美参考标准[2,3,20,30,38,39]。因此，本研究满足了指南对更大更广泛验证队列的要求，但并未完全满足对大量独立确诊FIP病例的单独要求[30]。回顾性单中心设计可能存在转诊和谱系效应，且性能依赖于平台和方案。此外，回顾性设计限制了对病例选择、辅助诊断信息完整性以及临床随访标准化的控制。尽管全部176份样本代表独特的猫，但最终替代诊断、两例假阴性积液的标准病例层面描述以及完整的细菌学数据并非始终可得。纳入了一例淋巴瘤性积液，但现有记录无法系统评估淋巴瘤或其他非脓毒性疾病作为假阳性结果的独立来源。脓毒症分析为单变量分析，脓毒症状态集中于渗出液，且脓毒性参考阳性积液和较少见积液类别的亚组精确度有限。此外，有意保留脓毒性积液以在临床现实的炎症条件下挑战特异性，尽管脓毒症样本数量仍然有限，尤其是在分子参考阳性猫中。因此，脓毒性积液的亚组估计应谨慎解读，主要作为关于特异性损失幅度的假设生成。
4.2. 建议
从实际诊断角度来看，FIP积液指数最恰当的定位可能是基于积液的早期分诊检测，而非独立的确定性诊断检测。对于具有相容临床症状和腔体积液的猫，可在初始诊断阶段整合常规细胞学、总蛋白、A/G比值和自动细胞计数数据。FIP积液指数阴性将大幅降低对FIP的怀疑，并可能支持优先考虑替代诊断，尤其是当细胞学和临床表现与FIP不一致时。相反，FIP积液指数阳性应增加诊断怀疑并促使直接进行FCoV检测。特别是，当组织可得时，应使用积液来源细胞涂片或细胞块上的免疫细胞化学和/或组织病变上的免疫组织化学检测FCoV抗原，作为更具特异性的确认证据。
未来研究应在疾病级诊断框架内评估FIP积液指数，采用组织病变中免疫组织化学法检测FCoV抗原和/或积液样本细胞块制备物中免疫细胞化学法检测FCoV抗原。前瞻性多中心验证应使用Sysmex XN-1000V上锁定的阈值7.54或更高、盲法判读以及独立裁定的参考标准（纳入组织病理学、FCoV抗原检测和临床随访）。应保留具有挑战性的炎症性疾病，进行系统性细菌培养，并在有条件时采用基质辅助激光解吸/电离飞行时间质谱，以探究不同细菌病因是否产生不同的细胞或蛋白反应及假阳性风险。还应评估不同分析仪和不同站点间的重现性。另一优先事项是开发并外部验证这种序贯诊断方法，最好将FIP积液指数纳入多变量临床-实验室评分中。在此类方法中，高灵敏度的FIP积液指数主要可在阴性时支持FIP排除，而指数阳性病例则需进行更具特异性的确证检测。该策略可在筛查阶段最大化灵敏度、在确证阶段最大化特异性，从而可能提高体内FIP诊断的可靠性。
5. 结论
预设的FIP积液指数阈值≥7.54在随后一个规模显著更大、诊断范围更广的队列中，以积液FCoV RT-qPCR为参考进行了验证。FIP积液指数在改良漏出液和渗出液表型中均保持极高的灵敏度，且连续判别能力优于单独ΔTNC或A/G，尽管相对于ΔTNC的增益有限。脓毒性炎症而非渗出液表型本身，是特异性的主要可识别局限。因此，FIP积液指数阴性提供了反对分子参考阳性的强有力证据，而阳性结果，尤其是在脓毒性积液中，应结合细胞学、细菌培养和直接FCoV检测进行解读。这些结果支持FIP积液指数作为Sysmex XN-1000V上稳健的辅助性积液生物标志物。
查看英文原文▾
The antemortem clinical and clinicopathological diagnosis of feline infectious peritonitis (FIP) remains challenging, particularly when effusions deviate from the classically protein-rich, low-to-moderate-cellularity phenotype. The FIP Effusion Index, defined as the ratio of delta total nucleated cells (ΔTNC) to the albumin-to-globulin (A/G) ratio, was developed and initially evaluated in 50 effusions. Subsequent European Advisory Board on Cat Diseases (ABCD) guidelines recommended evaluation in larger cohorts and in cats with confirmed FIP. This study retrospectively evaluated the prespecified threshold (≥7.54) in 176 independent feline effusion samples analysed on the Sysmex XN-1000V and classified by effusion FCoV RT-qPCR, without independent confirmation of FIP status. Relative to this molecular reference, the FIP Effusion Index yielded a sensitivity of 97.8% (95% confidence interval [CI], 92.4–99.7%), a specificity of 88.1% (95% CI, 79.2–94.1%) and an area under the receiver operating characteristic curve (AUC) of 0.962 (95% CI, 0.928–0.995). Discrimination exceeded that of ΔTNC alone (AUC, 0.947; ΔAUC, 0.014; p = 0.012) and A/G alone (AUC, 0.887; ΔAUC, 0.074; p = 0.012). In exudates, sensitivity remained 100% (95% CI, 80.5–100%) and specificity was 82.8% (95% CI, 64.2–94.2%). Among reference-negative effusions, false-positive classification was more frequent in septic than in non-septic samples (6/19 vs. 4/65; odds ratio, 7.04; 95% CI, 1.74–28.53; p < 0.01). These findings demonstrate that the prespecified FIP Effusion Index threshold retained very high sensitivity and strong discrimination for molecular-reference positivity across a broader spectrum of feline effusions. Future studies should evaluate the FIP Effusion Index within a sequential in vivo diagnostic algorithm, using negative results to support rule-out and positive results to prompt confirmation by FCoV antigen detection using immunocytochemistry and/or immunohistochemistry.
1. Introduction
Feline infectious peritonitis (FIP) is a systemic inflammatory disease that develops in a minority of cats infected with feline coronavirus (FCoV) and remains one of the most challenging diagnoses in feline medicine [1,2,3]. The availability of effective antiviral therapy has further increased the clinical importance of establishing an early and reliable diagnosis to permit prompt treatment [4,5,6,7,8,9]. When an effusion is present, its analysis should be prioritised, as the combination of cytology, biochemical assessment and direct detection of FCoV RNA or antigen provides substantially greater diagnostic information than peripheral blood testing [1,10,11,12]. Characteristic histopathological lesions together with intralesional FCoV antigen detection by immunohistochemistry (IHC) constitute the reference standard for definitive FIP diagnosis, although tissue sampling is invasive and may be impractical in clinically ill cats [1,2,10,11,12]. FCoV reverse-transcription quantitative polymerase chain reaction (RT-qPCR) on effusion samples is rapidly available and has shown high sensitivity and specificity, often reaching 100% in several studies [2,13,14,15,16,17,18,19,20,21]. Longstaff et al. [20] reported 85% sensitivity and 100% specificity in effusions, while Doenges et al. [21] reported 88.9% sensitivity and 100% specificity in cell-free body-cavity effusions, relative to disease classifications that included FCoV antigen detection by IHC. Although histopathology with intralesional FCoV antigen detection remains the reference standard, effusion RT-qPCR was selected as the molecular reference classification for this study because effusion sampling is less invasive than tissue biopsy, results are rapidly available and the assay is suitable for routine diagnostic workflows.
Classically, FIP-associated effusions are usually protein-rich, yellow to straw-coloured, viscous and sticky, often with fibrin or clot formation and frequently yielding a modified-transudative phenotype [3,22,23,24]. Cytologically, they usually contain a mixed pyogranulomatous population of non-degenerate neutrophils, macrophages or monocytes and few lymphocytes in a proteinaceous granular background [3,22,25,26,27]. Nevertheless, the biological spectrum is wider than this traditional description implies. Current FIP diagnostic guidelines recognise that total nucleated cell counts (TNCCs) can be substantially increased in FIP, including values exceeding 20 × 109/L, and that secondary bacterial peritonitis may coexist with FIP [1,2,3,28,29]. From a diagnostic perspective, this means that effusion phenotype should be interpreted cautiously: an exudative pattern does not exclude FIP, and severe inflammatory conditions unrelated to FIP may share some biochemical features commonly considered supportive of the disease [1,22,25,28].
Among inexpensive effusion-based diagnostic tests, the albumin-to-globulin (A/G) ratio and Rivalta test remain widely used [2,3,30]. Both, however, have important limitations. A low A/G ratio reflects the inflammatory protein profile of FIP but is not disease-specific, whereas a positive Rivalta reaction may occur in septic peritonitis and lymphoma-related effusions [2,17,31,32,33]. Diagnostic guidelines place a low effusion A/G ratio in the “more likely” rather than the “definitive” category, with values <0.4 increasing suspicion and >0.8 making FIP less likely [2,3,30,32]. The discordance between cellularity measurements obtained from the white blood cell differential (WDF) and white cell nucleated (WNR) analytical channels of the Sysmex XN-1000V provides a further diagnostic signal [1,11]. The parameter termed delta total nucleated cells (ΔTNC) has shown high diagnostic accuracy in feline effusions analysed on Sysmex haematology platforms [34,35].
The FIP Effusion Index was introduced in 2024 by our research group to integrate this cellular-channel phenomenon with the biochemical information contained in the A/G ratio [1]. In the derivation cohort of 50 effusions, a FIP Effusion Index threshold of ≥7.542 yielded 96.3% sensitivity and 100% specificity relative to effusion FCoV RT-qPCR status [1]. Importantly, that initial investigation excluded septic, unclear and inconclusive cytological samples, thereby evaluating a comparatively restricted diagnostic spectrum. The detailed 2026 European Advisory Board on Cat Diseases (ABCD) FIP guidelines subsequently highlighted the FIP Effusion Index as promising, specifically noting the need for evaluation in a larger number of samples and in more firmly confirmed FIP cases before routine recommendation [30].
The principal challenge in validating a diagnostic index is not merely to reproduce its performance in a population resembling the derivation cohort, but to determine whether a prespecified threshold remains informative when confronted with the heterogeneous inflammatory and neoplastic conditions encountered in routine practice. Re-optimising the threshold in each new dataset would inflate apparent performance and would not constitute a stringent validation. Accordingly, the present study retained ≥7.54 as the primary, prespecified clinical threshold and treated any data-driven threshold analysis as exploratory.
The primary objective was to assess the diagnostic performance of the FIP Effusion Index in a substantially larger, expanded-spectrum cohort of feline effusions using effusion FCoV RT-qPCR status as the molecular reference classification. Secondary objectives were to: (i) compare the continuous discrimination of the FIP Effusion Index with its individual components, ΔTNC and A/G; (ii) compare the prespecified FIP Effusion Index threshold with previously published thresholds for ΔTNC, A/G and the Rivalta test; (iii) determine whether sensitivity was retained in exudative effusions, which extend beyond the classically modified-transudative phenotype and may present a greater risk of misclassification; and (iv) investigate septic inflammation as a source of false-positive results. We hypothesised that the FIP Effusion Index would retain high sensitivity across effusion phenotypes, while severe septic inflammation would represent a biologically plausible limitation to specificity.
2. Materials and Methods
2.1. Study Design, Setting and Diagnostic Records
This retrospective, expanded-spectrum diagnostic-accuracy study was conducted using feline effusion samples submitted for routine diagnostic investigation to CEDIVET Veterinary Laboratories (Porto, Portugal) between 1 March 2025 and 1 March 2026. Eligible samples included effusions across different cytological and biochemical phenotypes, including modified transudates, exudates and septic inflammatory effusions. All performance estimates were calculated relative to effusion FCoV RT-qPCR classification and not to independently confirmed FIP status.
Effusion records were eligible for inclusion when a valid FCoV RT-qPCR result was available together with all laboratory measurements required to calculate the FIP Effusion Index, namely total nucleated cell counts from the WDF (TNCC-WDF) and WNR (TNCC-WNR) channels and the effusion A/G ratio. TNCC-WDF and TNCC-WNR measurements were obtained using the Sysmex XN-1000V analytical channels with Lysercell WDF (Cat. No. CV377552), Fluorocell WDF (Cat. No. AA325279), Lysercell WNR (Cat. No. DWX-50102) and Fluorocell WNR (Cat. No. M375584) reagents, respectively (Sysmex Corporation, Kobe, Japan). A Rivalta test result, effusion phenotype and septic classification were additionally required for the corresponding comparative and subgroup analyses.
Samples were submitted in ethylenediaminetetraacetic acid (EDTA; 4 mL; Cat. No. 454209, Greiner Bio-One, Kremsmünster, Austria) and serum tubes (4 mL; Cat. No. 454092, Greiner Bio-One, Kremsmünster, Austria) and processed within 18–24 h of collection, in accordance with the analytical protocol used in the original study [1]. Records were excluded when the molecular result was analytically invalid, the available sample volume was insufficient to complete the required analyses, or any measurement necessary to derive the primary index test, namely TNCC-WDF, TNCC-WNR, or the effusion A/G ratio, was unavailable. Specifically, samples with invalid amplification controls, insufficient volume, or missing TNCC-WDF, TNCC-WNR, or A/G ratio values were excluded. RT-qPCR results were interpreted according to the manufacturer’s qualitative criteria: samples with valid amplification controls and target amplification within the analytical run were classified as molecular-reference positive, whereas valid samples without target amplification were classified as molecular-reference negative. No separate borderline or indeterminate molecular category was defined for this study, and analytically invalid results were excluded from diagnostic-accuracy analyses. No records meeting the initial eligibility criteria were excluded for these reasons. Unlike the original derivation study, septic effusions were intentionally retained to evaluate the FIP Effusion Index under a broader spectrum of inflammatory conditions encountered in routine diagnostic practice.
The statistical unit was the individual effusion sample. All 176 samples represented independent observations from 176 individual cats, with no cat contributing more than one sample to the primary analysis. All eligible samples fulfilling the inclusion criteria during the predefined study period were included, and no imputation was performed for variables used in the primary diagnostic analyses. The study was reported according to the Standards for Reporting of Diagnostic Accuracy (STARD) [36].
2.2. Laboratory Analyses and Effusion Classification
Cytological examination was performed on direct smears, squash preparations or cytocentrifuged specimens, as appropriate, stained with a Romanowsky-type stain using a RAL Stainer (CellaVision RAL Diagnostics, Martillac, France) [1].
Effusions were classified according to total protein concentration, cellularity and cytological characteristics. The traditional framework categorised body fluids as transudates, modified transudates or exudates [1,37]. The previously established criteria comprised total protein < 2.5 g/dL and TNCC < 1500 cells/μL for transudates; total protein 2.5–7.5 g/dL and TNCC 1000–7000 cells/μL for modified transudates; and total protein > 3.0 g/dL and TNCC > 7000 cells/μL for exudates [1].
Total protein and albumin concentrations were measured using a Beckman Coulter AU680e analyser (Beckman Coulter, Cassina de’ Pecchi, Italy) by the biuret (Cat. OSR6132, Beckman Coulter®, Brea, CA, USA) and bromocresol green methods (Cat. OSR6102, Beckman Coulter®, Brea, CA, USA), respectively. Globulin concentration was calculated as total protein minus albumin, and the A/G ratio as albumin divided by globulin. The Rivalta test was performed and interpreted according to the previously described protocol [1,31,33].
Nucleated-cell measurements were obtained using the Sysmex XN-1000V haematology analyser (Sysmex Corporation, Kobe, Japan) in whole-blood mode. ΔTNC was operationally defined as the ratio (1):
[内容请查看原文]
ΔTNC=TNCC-WDFTNCC-WNR where TNCC-WDF and TNCC-WNR represent the total nucleated cell counts generated by the white blood cell differential and white cell nucleated channels, respectively [1,34,35].The anatomical site of the sampled effusion was recorded as abdominal, pleural or pericardial. This sample-level classification does not imply that effusion was confined to a single cavity in each cat. Cytological evidence of intracellular bacteria within an inflammatory effusion supported classification as septic, and microbiological culture was considered corroborative when available.
For the prespecified phenotype analysis, modified transudates were compared with effusions belonging to the exudative family. Less frequent phenotypes, comprising chylous, transudative and haemorrhagic effusions, remained included in the overall analysis.
2.3. Molecular Analysis and Reference Classification
Viral RNA was extracted from effusion samples using the Promega AX9760 (Promega Corporation, Madison, WI, USA) custom nucleic-acid extraction protocol for whole blood, adapted for body-fluid samples and implemented on the KingFisher™ Flex system (Thermo Fisher Scientific, Waltham, MA, USA), following the previously described laboratory procedure [1]. FCoV RNA was detected using the NZYtech Feline Coronavirus RT-qPCR Kit (Cat. MD05601, NZYtech®, Lisbon, Portugal) on a QuantStudio™ 5 Real-Time PCR System (Thermo Fisher Scientific, Waltham, MA, USA), according to the manufacturer’s instructions. Thermal cycling comprised reverse transcription at 50 °C for 20 min, polymerase activation at 95 °C for 2 min, and 40 cycles of denaturation at 95 °C for 5 s and annealing/extension at 60 °C for 30 s [1]. Each reaction included an internal extraction control, and each analytical run included appropriate positive and negative controls. A molecular result was considered analytically valid only when the internal extraction control amplified appropriately, the positive control showed amplification and no target amplification was observed in the negative control. No study-specific quantification-cycle cut-off was introduced. Valid samples were classified according to the manufacturer’s qualitative criteria, with target amplification at a quantification cycle (Cq) < 40 and the appropriate sigmoidal profile considered positive. Valid samples without target amplification were classified as negative. Invalid results were excluded from the diagnostic-accuracy analysis. The qualitative positivity criterion does not distinguish low from high viral loads. Late amplification meeting the same validity and positivity criteria was not assigned a separate study-specific category.
The qualitative effusion FCoV RT-qPCR result constituted the molecular reference classification. Group allocation was based solely on FCoV RT-qPCR and was presumptive with respect to FIP status. Detection of FCoV RNA in effusion is not an error-free reference standard for definitive FIP; diagnostic-performance estimates therefore describe discrimination and classification relative to molecular status, rather than absolute disease status [38,39,40]. Although histopathology with intralesional FCoV antigen detection remains the reference standard, effusion RT-qPCR was selected as the molecular reference classification for this study because effusion sampling is less invasive than tissue biopsy, results are rapidly available and the assay is suitable for routine diagnostic workflows. This choice does not establish independent disease confirmation. The assay detects FCoV without distinguishing feline enteric coronavirus (FECV) from FIP-associated variants (FIPV), and a negative effusion RT-qPCR result alone cannot exclude FIP [30].
2.4. FIP Effusion Index and Prespecified Thresholds
The FIP Effusion Index was calculated according to Equation (2) originally proposed by Lopes et al. [1]:
[内容请查看原文]
FIP Effusion Index=ΔTotal Nucleated Cells (ΔTNC)Albumin/Globulin (A/G) ratioThe FIP Effusion Index was calculated as a continuous variable before application of any diagnostic threshold. The primary index test was evaluated at the prespecified clinical threshold of ≥7.54, representing the rounded form of the ≥7.542 threshold derived in the previous study independently of the present dataset and subsequently cited in the 2026 ABCD guidelines [1,30]. The historical unrounded threshold of ≥7.542 was evaluated separately as a sensitivity analysis and was not used to redefine the primary operating point.
Previously published comparator thresholds were ΔTNC ≥ 4.903, ΔTNC ≥ 2.081 and A/G ≤ 0.56 [1]. Rivalta test positivity was evaluated as a binary comparator.
The primary diagnostic outcomes were sensitivity and specificity. Secondary measures included overall accuracy, positive and negative likelihood ratios (LR+ and LR−), and the diagnostic odds ratio (DOR). Throughout this study, sensitivity, specificity, accuracy, likelihood ratios and true/false-positive or -negative classifications are defined relative to effusion FCoV RT-qPCR, not independently established FIP status.
2.5. Statistical Analysis
Continuous variables were summarised as median and interquartile range (IQR), and categorical variables as counts and percentages. Differences between FCoV RT-qPCR-positive and -negative effusions were assessed using two-sided Mann–Whitney U tests, with rank-biserial correlation reported as an effect-size measure.
Diagnostic performance at the prespecified thresholds was evaluated from two-by-two contingency tables. Sensitivity, specificity and accuracy were calculated with exact 95% confidence intervals (CIs), together with LR+, LR− and DORs. Continuous discrimination of the FIP Effusion Index, ΔTNC and A/G ratio was assessed using receiver operating characteristic (ROC) curves and areas under the curve (AUCs). A/G was directionally reversed because lower values support FIP. AUCs were compared using paired DeLong tests [41], with Holm adjustment applied to the three pairwise comparisons.
Fixed-threshold differences in sensitivity and specificity were assessed using exact McNemar tests with Holm adjustment. Prespecified subgroup analyses compared modified transudates with exudates and septic with non-septic effusions; abdominal and pleural effusions were examined exploratorily. Independent proportions were compared using Fisher’s exact test.
The Youden index was used only for exploratory assessment of threshold concordance. AUC and exploratory threshold stability were evaluated using 3000 stratified bootstrap resamples preserving reference-class proportions. No post hoc power calculation was performed. All tests were two-sided, with p < 0.05 considered statistically significant [42].
Statistical analyses were performed using JMP® version 18.0.0 (SAS Institute Inc., Cary, NC, USA), DATAtab® (numiqo e.U., Graz, Austria, 2026), MedCalc® Statistical Software version 20.006 (MedCalc Software Ltd., Ostend, Belgium), and R 4.6.1 (R Foundation for Statistical Computing, Vienna, Austria).
3. Results
3.1. Study Population
The analysis dataset comprised 176 effusion records, of which 92 (52.3%) were FCoV RT-qPCR-positive and 84 (47.7%) were negative. Modified transudates were the predominant phenotype (117/176; 66.5%), followed by exudates (46/176; 26.1%); 13 records (7.4%) comprised seven chylous, five transudative and one haemorrhagic effusion. Septic inflammation was recorded in 23/176 (13.1%) observations. The sampled effusions comprised 118 abdominal (67.0%), 55 pleural (31.3%) and three pericardial (1.7%) effusions. Anatomical location and sex were available for all records. Signalment and principal fluid characteristics are summarised in Table 1.
3.2. Continuous Effusion Markers According to Reference Classification
Molecular-reference-positive effusions had higher total protein and globulin concentrations and lower A/G ratios than reference-negative effusions (Table 2). The largest univariable effect sizes were observed for the FIP Effusion Index (rank-biserial r = 0.923) and ΔTNC (r = 0.894). Median FIP Effusion Index values were 45.90 (IQR 25.41–94.81) in reference-positive effusions and 1.21 (0.86–2.14) in reference-negative effusions (p < 0.001; rank-biserial r = 0.923). Corresponding median ΔTNC values were 19.09 (9.82–37.00) and 0.98 (0.87–1.17), respectively (both p < 0.001). Globulin (r = 0.779) and A/G (r = −0.775) also showed substantial separation. In contrast, TNCC-WDF alone provided little separation (median 1.987 vs. 2.800 × 109/L; p = 0.090; r = −0.148). Figure 1 displays the FIP Effusion Index distribution at the locked threshold.
3.3. Primary Validation of the Prespecified FIP Effusion Index Threshold
Application of the independently established threshold of ≥7.54, without optimisation in this dataset, classified 90/92 reference-positive and 74/84 reference-negative effusions correctly (TP/FP/FN/TN, 90/10/2/74). Sensitivity was 97.8% (95% CI, 92.4–99.7%), specificity was 88.1% (95% CI, 79.2–94.1%) and overall accuracy was 93.2% (95% CI, 88.4–96.4%). LR+ was 8.22 (95% CI, 4.59–14.71) and LR− was 0.025 (95% CI, 0.006–0.097), corresponding to a DOR of 333 (95% CI, 70.7–1567.4).
Use of the exact historical threshold of ≥7.542 rather than its rounded clinical form resulted in one additional false-negative classification: sensitivity was 96.7%, while specificity remained 88.1%. The clinical threshold of ≥7.54 was therefore retained for all primary and subgroup analyses.
3.4. Continuous Discrimination and Incremental Value
The continuous FIP Effusion Index showed excellent discrimination, with an AUC of 0.962 (95% CI, 0.928–0.995). The AUCs of ΔTNC and A/G were 0.947 (95% CI, 0.910–0.984) and 0.887 (95% CI, 0.837–0.937), respectively (Figure 2). Because ROC AUC evaluates the continuous marker, the ΔTNC AUC is independent of the fixed cut-offs. The operating characteristics at ΔTNC ≥ 4.903 and ≥2.081 are reported separately in Table 3. The complementary two-dimensional relationship between ΔTNC and A/G at the fixed FIP Effusion Index threshold is shown in Figure 3. In paired DeLong analysis, the FIP Effusion Index exceeded ΔTNC by 0.014 AUC units (95% CI, 0.004–0.025; p < 0.01; Holm-adjusted p = 0.012) and A/G by 0.074 (95% CI, 0.024–0.125; p < 0.01; adjusted p = 0.012). ΔTNC also exceeded A/G (ΔAUC, 0.060; 95% CI, 0.004–0.115; p = 0.034; adjusted p = 0.034). Differences were calculated from unrounded AUC estimates. Thus, combining ΔTNC with A/G produced a statistically demonstrable, although quantitatively modest, improvement over ΔTNC alone.
3.5. Paired Fixed-Threshold Comparisons
At fixed cut-offs, the FIP Effusion Index correctly identified six reference-positive effusions missed by ΔTNC ≥ 4.903, while no reference-positive effusion was correctly identified by ΔTNC and missed by the Index. The exact McNemar p value was 0.031. However, this difference did not remain significant after Holm adjustment across the prespecified fixed-threshold comparison family (adjusted p = 0.125). In contrast, the sensitivity advantage of the FIP Effusion Index over A/G ≤ 0.56 remained significant (adjusted p < 0.01). The principal fixed-threshold advantage over the Rivalta test was specificity: among reference-negative records, 17 were correctly classified by the FIP Effusion Index but positive by the Rivalta test, with no discordance in the opposite direction (exact p < 0.001; adjusted p < 0.001).
3.6. Performance Beyond Modified Transudates
The prespecified threshold retained high sensitivity in both major effusion phenotypes. Among 117 modified transudates, sensitivity was 97.3% (95% CI, 90.7–99.7%) and specificity was 92.9% (95% CI, 80.5–98.5%), with an AUC of 0.970. Among 46 exudates, all 17 reference-positive records were FIP Effusion Index-positive, giving an observed sensitivity of 100% (95% CI, 80.5–100%); specificity was 82.8% (95% CI, 64.2–94.2%) and the AUC was 0.947 (95% CI, 0.876–1.000). Sensitivity did not differ between modified transudates and exudates (Fisher p = 1.000). Specificity was 10.1 percentage points higher in modified transudates than in exudates (difference, +10.1 percentage points; 95% CI, −5.2–28.0; Fisher p = 0.258). These results provide no evidence that exudative classification causes loss of sensitivity. Within the exudative samples, no misclassification occurred among 26 non-septic effusions (14 reference-positive and 12 reference-negative), whereas 5/17 reference-negative septic exudates were FIP Effusion Index-positive. The septic-exudate subgroup nevertheless retained an AUC of 0.941, suggesting that the continuous marker remained strongly informative even though a subset of severe inflammatory effusions crossed the fixed threshold. Given the small subgroup sizes and zero cells in some strata, these estimates should be interpreted through their confidence intervals rather than as evidence of perfect subgroup classification.
3.7. Septic Inflammation and False-Positive Classification
Septic inflammation emerged as the principal identifiable context associated with reduced specificity. Across all molecular reference-negative effusions, false-positive FIP Effusion Index results occurred in 6/19 (31.6%) septic records compared with 4/65 (6.2%) non-septic records. Specificity was therefore 68.4% in septic effusions and 93.8% in non-septic effusions, corresponding to a difference of −25.4 percentage points (95% CI, −48.1 to −7.1; Fisher p < 0.01). The odds of false-positive classification were approximately seven-fold higher in septic than in non-septic reference-negative effusions (OR, 7.04; 95% CI, 1.74–28.53). The distribution of observations according to molecular reference and septic status is shown in Figure 3.
Notably, septic inflammation did not invariably result in an increased FIP Effusion Index (Table 4). A representative FCoV RT-qPCR-negative septic exudate correctly classified below the prespecified threshold is shown in Figure 4 and Figure 5. Despite marked inflammatory cytological findings, this effusion had a ΔTNC of 1.00, an A/G ratio of 1.19 and a resulting FIP Effusion Index of 0.84. This example illustrates that even marked septic suppurative inflammation does not necessarily reproduce the inter-channel discordance associated with a high FIP Effusion Index.
3.8. Exploratory Threshold Stability and Bootstrap Analysis
Exploratory Youden analysis in the full validation cohort showed that the prespecified threshold of 7.54 achieved the maximum Youden J (0.859), with a sensitivity of 97.8% and a specificity of 88.1%. The algorithm-derived optimal threshold was 6.300, which yielded identical sensitivity, specificity, and Youden J to the prespecified threshold. Across 3000 stratified bootstrap resamples, the median algorithm-derived optimal threshold was 6.300 (IQR, 6.171–6.362), with a broad 2.5th–97.5th percentile range of 3.62–17.42, indicating appreciable finite-sample instability in data-driven cut-off selection. The bootstrap AUC distribution was more stable (median, 0.964; percentile range, 0.924–0.990). These exploratory findings show that the prespecified threshold achieves the maximum Youden performance in the validation cohort, while the bootstrap results illustrate the instability of selecting a single data-driven optimal threshold; they do not retroactively validate or re-derive the prespecified cut-off (Figure 6).
4. Discussion
This study was designed not to redefine the cut-off, but to determine whether the previously established FIP Effusion Index and its prespecified threshold [1] remained informative in a temporally distinct, substantially larger cohort spanning a broader spectrum of effusions. Four clinically important findings emerged. First, the threshold of ≥7.54 retained very high sensitivity (97.8%; 95% CI, 92.4–99.7%), strong overall discrimination (AUC, 0.962; 95% CI, 0.928–0.995) and a DOR of 333 (95% CI, 70.7–1567) in a cohort more than three times larger than the original derivation series [1]. Second, the composite FIP Effusion Index discriminated reference-positive from reference-negative effusions better than either ΔTNC or A/G alone, although the gain over ΔTNC was quantitatively modest. Third, high sensitivity was retained among exudates, indicating that an exudative phenotype did not abolish the analytical signal captured by the Index. Fourth, septic inflammation was the most evident context associated with reduced specificity. These observations define both the potential clinical value and a clinically recognisable limitation of the FIP Effusion Index without equating molecular-reference positivity with independently confirmed FIP.
The validation strategy is a major strength of the present work. Both the FIP Effusion Index formula and the clinical threshold were established in an independent derivation cohort and were locked before the present data were analysed [1]. This distinction matters because post hoc threshold selection can produce optimistic estimates that fail to generalise [43,44]. Temporal separation, inclusion of all eligible cats during the predefined period and deliberate retention of inflammatory competitors therefore constitute a more stringent test of transportability. Exploratory Youden analysis showed that the locked threshold of 7.54, subsequently reproduced in ABCD guidelines [30], achieved the same maximum Youden J as the algorithm-derived optimal threshold of 6.300, with identical sensitivity and specificity, thereby supporting concordance with the previously established threshold rather than its retrospective re-derivation. The median bootstrap algorithm-derived optimal threshold was 6.300, with a broad percentile range (3.62–17.42), further illustrating the instability of data-driven cut-off selection in finite validation samples and why an established clinical threshold should not be repeatedly recalibrated.
The present findings extend, rather than simply reproduce, the original FIP Effusion Index study [1] and the earlier ΔTNC literature [10,34,35]. In the 50-effusion derivation cohort, the ≥7.542 threshold yielded 96.3% sensitivity and 100% specificity [1]. In this study, sensitivity was similar, but specificity was lower at 88.1% (95% CI, 79.2–94.1%). These point estimates should not be interpreted as directly interchangeable because the present sample was larger and included septic, exudative and otherwise diagnostically challenging fluids that had been restricted or excluded during derivation. Earlier XT-2000iV studies established ΔTNC as a highly discriminating effusion marker [10,34,35], whereas subsequent comparison of the XT-2000iV and XN-1000V demonstrated systematic inter-analyser differences despite broadly similar diagnostic classification [11]. Thus, the current results strengthen the evidence for the FIP Effusion Index specifically on the XN-1000V under the stated pre-analytical and analytical conditions. Consequently, neither ΔTNC values nor the FIP Effusion Index threshold should be assumed to be transferable to another analyser without platform-specific validation.
The paired analyses help explain why integration is useful. ΔTNC, operationally the ratio of TNCC-WDF to TNCC-WNR, captures inter-channel discordance arising from distinct reagent and measurement environments, whereas A/G describes the distribution of albumin relative to globulins within the inflammatory protein milieu. Protein-rich FIP effusions, monocyte/macrophage-driven inflammation and altered vascular permeability are well documented [45,46,47], and fibrin- or protein-associated cellular aggregation or trapping in one analytical channel has been proposed as a contributor to ΔTNC [10,11,34]. The precise physicochemical mechanism on the XN-1000V has not been experimentally resolved, however, and this explanation should remain a hypothesis rather than be presented as established analyser behaviour. By combining partially complementary cellular/analytical and biochemical information, the FIP Effusion Index increased AUC by 0.014 over ΔTNC (95% CI, 0.004–0.025; Holm-adjusted p = 0.012) and by 0.074 over A/G (95% CI, 0.024–0.125; adjusted p = 0.012). The gain over ΔTNC was statistically supported but quantitatively modest. At fixed thresholds, the FIP Effusion Index recovered six reference-positive effusions missed by ΔTNC ≥ 4.903, but this sensitivity comparison did not remain significant after multiplicity adjustment (adjusted p = 0.125). The broader principle that ratios integrating distinct effusion measurements may improve diagnostic discrimination is also supported by previous work evaluating the effusion lactate dehydrogenase (LDH)/TNCC ratio, although that marker and the present FIP Effusion Index are not analytically or clinically interchangeable [48].
Diagnostic performance in exudates is a central advance of this validation. Effusive FIP is conventionally associated with a protein-rich, low-to-moderate-cellularity modified transudate, but both current guidelines and series of feline effusions demonstrate substantial overlap between cytological categories and aetiologies [12,30,49,50,51]. Traditional transudate, modified-transudate and exudate labels describe fluid characteristics; they are neither diagnostic nor disease-specific. All 17 reference-positive exudates were FIP Effusion Index-positive, corresponding to an observed sensitivity of 100%, but the 95% CI of 80.5–100% and the small denominator preclude any claim of perfect performance. Specificity in exudates was 82.8% (95% CI, 64.2–94.2%), and the data provided no evidence that sensitivity differed from that in modified transudates. Accordingly, exudative classification alone should neither exclude FIP from the differential diagnosis nor invalidate interpretation of the Index. Conversely, an exudative FIP Effusion Index-positive effusion should not be presumed to represent FIP, particularly when cytological evidence of bacterial inflammation is present.
The lower overall specificity compared with derivation [1] is therefore scientifically informative. Diagnostic performance varies with case mix, competing disorders and the distribution of disease severity, even when the test itself is unchanged [44,52]. In this cohort, specificity was 93.8% (95% CI, 85.0–98.3%) among non-septic effusions but 68.4% (95% CI, 43.5–87.4%) among septic effusions. The broader cohort is consequently a harder and more clinically relevant challenge than the restricted derivation set, but it should not be assumed to represent every intended-use population. The results locate much of the observed loss of specificity within a defined inflammatory context rather than supporting a general loss of discrimination by the FIP Effusion Index.
Among the 84 reference-negative effusions, false-positive classification occurred in 6/19 (31.6%) septic samples and 4/65 (6.2%) non-septic samples, an absolute specificity difference of −25.4 percentage points (95% CI, −48.1 to −7.1) and an unadjusted odds ratio of 7.04 (95% CI, 1.74–28.53). This establishes an association, not causation or independence from effusion phenotype. The analysis was univariable, and septic status was concentrated among exudates. Overlap in globulin enrichment, low A/G, fibrin content and cellular–reagent behaviour offers plausible explanations, but these mechanisms were not directly measured [40,45,46,53,54]. Equally, sepsis did not invariably produce a high FIP Effusion Index: the representative septic exudate had marked suppurative inflammation but a ΔTNC of 1.00, A/G of 1.19 and FIP Effusion Index of 0.84. Severe cytological inflammation alone is therefore insufficient to reproduce the complete signature. Clinically, an FIP Effusion Index-positive effusion containing degenerate neutrophils and intracellular bacteria warrants increased caution and integration with cytology, bacterial culture where available, direct FCoV testing and the full clinical context [30,55]. Septic effusions should not simply be excluded from interpretation, and bacterial inflammation should not automatically be treated as mutually exclusive with FIP because secondary bacterial complications can occur [30].
Likelihood ratios translate these operating characteristics more usefully than sensitivity and specificity alone [56]. The LR− of 0.025 (95% CI, 0.006–0.097) was particularly strong in this cohort. For an assumed 50% pre-test probability of effusion RT-qPCR positivity, the pre-test odds are 1.0. Multiplying by 0.025 gives post-test odds of 0.025 and a post-test probability of approximately 2.4%. A negative FIP Effusion Index result would therefore markedly reduce the probability of molecular-reference positivity under that assumption. By contrast, at an assumed pre-test probability of 10%, the LR+ of 8.22 changes odds from 0.111 to 0.913, equivalent to a post-test probability of approximately 47.7%: a substantial increase, but not a definitive result. These examples concern the molecular reference used here, not histopathologically confirmed FIP.
The FIP Effusion Index should also be positioned alongside, rather than in opposition to, conventional tests. The Rivalta test and A/G remain inexpensive, rapid and widely accessible, and neither requires the specified haematology platform [2,3,30,55]. In the present paired data, the Rivalta test retained high sensitivity (96.7%) but lower specificity (67.9%). The FIP Effusion Index correctly classified 17 reference-negative effusions that were Rivalta-positive, with no discordance in the opposite direction. Low A/G is common in confirmed FIP but also accompanies other inflammatory states, limiting its stand-alone specificity [17,45,55]. Where TNCC-WDF, TNCC-WNR and effusion biochemistry are already available, their integration therefore provides greater discrimination than either component alone and may improve specificity over the Rivalta test. Direct FCoV antigen assays on effusion macrophages provide different information but have also shown variable operating characteristics across studies [29,57,58,59]. A recent diagnostic clinical trial of multiplex fluorescent immunocytochemistry likewise concluded that the assay was complementary rather than definitive [59]. Collectively, these data favour a multimodal diagnostic workflow over replacement of clinical reasoning by any single effusion test.
The choice of reference classification sets an essential boundary around the present inferences. Effusion FCoV RT-qPCR is clinically valuable because viral RNA is often abundant in FIP-associated effusions [60]. Previous studies reported sensitivities of 85.0% and 88.9% and specificities of 100% relative to their respective disease classifications [20,21]. Nevertheless, RT-qPCR detects FCoV RNA rather than the defining tissue lesion and is not a perfect reference standard [20,21,24,30,55]. Immunocytochemical and immunofluorescent detection of viral antigen in effusion macrophages is also imperfect [29,57,58,59]. The 2026 ABCD guidance identified two unresolved evidence needs for the FIP Effusion Index: evaluation in a larger sample and evaluation in cats with confirmed FIP [30]. This study substantially addresses the first through temporal validation in 176 cats, broader fluid phenotypes, inclusion of septic and exudative competitors and application of the original locked threshold. It does not fully address the second because classification was not based on an independently adjudicated disease diagnosis. The next step should therefore move beyond further analytical validation towards prospective multicentre disease-level evaluation of the FIP Effusion Index within a sequential multivariable diagnostic framework, in which its high sensitivity supports initial rule-out of FIP, while positive cases undergo more specific confirmatory testing, particularly FCoV antigen detection by immunocytochemistry in effusion-derived cell blocks and/or immunohistochemistry in tissue lesions. Such an approach could maximise sensitivity at the screening stage and specificity at the confirmatory stage and could thereby improve the reliability of in vivo FIP diagnosis.
4.1. Limitations
Several limitations require explicit consideration. Most importantly, the reference classification was effusion FCoV RT-qPCR rather than histopathology with intralesional FCoV antigen demonstration. Although effusion RT-qPCR is highly informative in the appropriate clinical context, it is not a perfect reference standard for definitive FIP diagnosis [2,3,20,30,38,39]. Consequently, this study addresses the guideline request for a larger and broader validation cohort but does not fully satisfy the separate requirement for a large series of independently confirmed FIP cases [30]. The retrospective, single-centre design may be subject to referral and spectrum effects, and performance is platform- and protocol-dependent. Also, the retrospective nature of the design limited control over case selection, completeness of ancillary diagnostic information and standardisation of clinical follow-up. Although all 176 samples represented unique cats, final alternative diagnoses, standardised case-level descriptions of the two false-negative effusions and complete bacteriological data were not consistently available. One lymphomatous effusion was included, but the available records did not permit systematic assessment of lymphoma or other non-septic disorders as independent sources of false-positive results. The sepsis analysis was univariable, septic status was concentrated among exudates and subgroup precision was limited for septic reference-positive effusions and less frequent fluid categories. In addition, septic effusions were deliberately retained to challenge specificity under clinically realistic inflammatory conditions, although the number of septic samples remained modest, particularly among molecular-reference-positive cats. Therefore, subgroup estimates for septic effusions should be interpreted cautiously and primarily as hypothesis-generating with respect to the magnitude of specificity loss.
4.2. Recommendations
From a practical diagnostic perspective, the FIP Effusion Index may be most appropriately positioned as an early effusion-based triage test rather than as a stand-alone definitive diagnostic test. In cats with compatible clinical signs and cavitary effusion, routine cytology, total protein, A/G ratio and automated cell-count data could be integrated at an initial diagnostic stage. A negative FIP Effusion Index would substantially reduce suspicion of FIP and may support prioritisation of alternative diagnoses, particularly when cytology and clinical findings are discordant with FIP. Conversely, a positive FIP Effusion Index should increase diagnostic suspicion and prompt direct FCoV testing. In particular, FCoV antigen detection by immunocytochemistry on effusion-derived cell preparations or cell blocks and/or immunohistochemistry on tissue lesions, when tissue is available, should be used as more specific confirmatory evidence.
Future studies should evaluate the FIP Effusion Index within a disease-level diagnostic framework using FCoV antigen detection by immunohistochemistry in tissue lesions and/or immunocytochemistry in cell-block preparations obtained from effusion samples. Prospective multicentre validation should use the locked threshold of 7.54 or higher on the Sysmex XN-1000V, blinded interpretation and an independently adjudicated reference incorporating histopathology, FCoV antigen detection and clinical follow-up. Challenging inflammatory conditions should be retained, with systematic bacterial culture and, where available, matrix-assisted laser desorption/ionisation time-of-flight mass spectrometry to investigate whether different bacterial aetiologies produce distinct cellular or protein responses and false-positive risks. Reproducibility across analysers and sites should also be assessed. A further priority is the development and external validation of this sequential diagnostic approach, ideally incorporating the FIP Effusion Index within a multivariable clinical–laboratory score. Within such an approach, the highly sensitive FIP Effusion Index could primarily support FIP rule-out when negative, whereas Index-positive cases would undergo more specific confirmatory testing. This strategy could maximise sensitivity at the screening stage and specificity at the confirmatory stage and could thereby improve the reliability of in vivo FIP diagnosis.
5. Conclusions
The prespecified FIP Effusion Index threshold of ≥7.54 was validated against effusion FCoV RT-qPCR in a subsequent, substantially larger and diagnostically broader cohort. The FIP Effusion Index retained very high sensitivity across modified-transudative and exudative phenotypes and showed greater continuous discrimination than either ΔTNC or A/G alone, although the gain over ΔTNC was modest. Septic inflammation, rather than exudative phenotype per se, was the principal identifiable limitation to specificity. Accordingly, a negative FIP Effusion Index provides strong evidence against molecular-reference positivity, whereas a positive result, particularly in a septic effusion, should be interpreted alongside cytology, bacterial culture and direct FCoV testing. These results support the FIP Effusion Index as a robust adjunctive effusion biomarker on the Sysmex XN-1000V.