Smallholder dairy farming constitutes the predominant form of livestock production in low- and middle-income countries (LMICs), contributing significantly to nutrition, household income, and livelihoods. However, productivity in these systems is hampered by several factors including poor reproductive efficiency, inadequate nutrition, infectious diseases, heat stress, and suboptimal management practices ( Bang et al., 2022; Banda et al., 2021). Effective reproductive management, particularly oestrus detection, is critical for improving reproductive indices such as pregnancy rate, heat detection rate, days open, and inseminations per conception ( Crowe et al., 2018; Endo 2022).

In many LMIC settings, oestrus detection is based on visual observation. This approach often fails due to farmers’ limited capacity to detect signs of oestrus accurately, resulting in delayed or missed inseminations ( Alawneh et al., 2006; Adenuga et al., 2020). Oestrus expression is modulated by physiological variables such as parity and milk yield, and environmental stressors like heat and mobility ( Wrzecińska et al., 2021; Szalai et al., 2025). For example, lameness and negative energy balance are known to reduce the intensity of oestrus behaviours, such as increased locomotion or standing heat ( Cook et al., 2007; Walker et al., 2008; Das et al., 2016; Polsky & Von Keyserlingk, 2017).

Oestrus detection tools (ODTs) have been developed to automate or assist with detection based on physiological or behavioral markers. These tools function as screening tests, rather than diagnostic tools, because they predict the likelihood of a cow being in oestrus without providing definitive confirmation ( Firk et al., 2002; Chung et al., 2013; Adenuga et al., 2020). Performance evaluation of ODTs, therefore, hinges on sensitivity (ability to identify true positives), specificity (ability to rule out non-oestrus), and robustness under diverse management systems ( Cavalieri et al., 2003; Brassel et al., 2018).

This study was the second of three sequential experiments, conducted separately and under varied experimental and field conditions, but designed in a progressive way (part of PhD project Muasa B., 2020) and aimed to assess the agreement, effectiveness, and robustness of three ODTs – CowAlert ^® (activity monitor), Estrotect™ (scratch card), and P4 Rapid™ (progesterone test) under the tropical conditions of a smallholder dairy system in Kenya. Specific objectives were: (1) to determine the agreement among ODTs, (2) to evaluate their sensitivity and specificity in detecting true oestrus events, and (3) to explore how intrinsic factors like parity, milk yield, body weight, and days in milk influence their effectiveness.

This represents the first study, to our knowledge, from a low- and middle-income country aimed at assessing the level of agreement among oestrus detection tools, evaluating their effectiveness, and identifying factors that may influence their performance. Notably, prior infection with FMD emerged as a potential confounding variable, possibly affecting cow performance in subtle or unrecognised ways during the study. Previous research has indicated that FMD can impair reproductive function in cattle ( Chaters et al., 2018).

A total of 130 oestrus events were identified in 51 cows using a temporal progesterone profiling approach. Notably, only 38% of the recorded oestrous cycles fell within the 17 to 28-day range, which deviates from the typical cycle length of 18 to 24 days as described by Forde et al. (2011). This apparent prolongation of cycle length may be attributed to undetected oestrus events by the progesterone-based method, resulting in an overestimation of cycle duration. However, other studies utilizing sensor-based technologies have reported oestrous cycle lengths ranging from 15 to 26 days ( Cushman et al., 2024).

The results on agreement between the oestrus detection tools (ODTs) revealed statistically significant differences in the detection outcomes when P4 Rapid™ was compared with CowAlert ^® and Estrotect™, indicating poor concordance between P4 Rapid™ and the other two tools. This suggests that the tools do not consistently identify the same oestrus events, likely due to their reliance on different physiological or behavioral indicators. Supporting this observation, Holman et al. (2011) reported that scratch cards had markedly lower sensitivity (36%) for detecting true oestrus compared to activity meters, neck collars, and visual observation, all of which showed sensitivities of approximately 60%. These findings align with the expectation that ODTs designed to measure distinct signs or predictors of oestrus vary in performance depending on their robustness and the predictive value of the indicator being measured. Oestrus signs may be classified as primary, such as the classic ‘standing to be mounted’ behavior, or secondary, including increased restlessness, mounting other cows, clear mucous discharge, dirty flanks, vulvar swelling and redness, as well as reduced feed intake and milk yield. Various ODTs target the primary sign of standing oestrus, such as tail paint, pressure-sensitive devices, and electronic mount detectors. In contrast, activity-based indicators reflect secondary signs, which are more susceptible to external factors like herd dynamics, housing systems, and environmental or management conditions.

The effectiveness of the three ODTs was assessed, with P4 Rapid demonstrating the highest sensitivity at 98% and a specificity of 51%. This high sensitivity aligns with findings by Ingenhoff et al. (2016), who reported a sensitivity of 90.1%, and Waldmann and Raud (2016), who found a sensitivity of 95%. In commercial dairy operations, where maximizing the identification of cows is likely in oestrus is critical, tools with high sensitivity are particularly valuable. While specificity which is correctly identifying non-oestrus events is less of a priority in such settings, the performance of P4 Rapid indicates it is the most advantageous tool under these conditions.

The reliability of CowAlert ^® as an ODT was compromised by recurrent system instability in this production setting. Over the 18-month study period, 11 outages were recorded, each lasting between 8 and 120 hours, with an average duration of less than 48 hours. These disruptions often triggered false heat alerts, with a notable instance where a 120-hour outage led to 48 cows being incorrectly flagged as in oestrus. To mitigate this, all oestrus events detected within and up to 24 hours after an outage were excluded from the analysis, thereby reducing the dependability of CowAlert ^® for accurate heat detection.

In this study, cows spent most of the day grazing in open fields, averaging approximately 5,104 steps per day. Given an average pasture stride length of 1.5 meters ( Alsaaod et al., 2017), this translates to an estimated distance of about 7.6 km per day. Oestrus behaviours tend to be less evident in grazing cows than in those kept indoors ( Palmer et al., 2012). The CowAlert ^® system uses algorithms that compute a rolling average of activity, often based on the previous 24 hours. For example, McGowan et al. (2007) analysed a 24-hour moving average and calculated its rate of change using regression on the prior 10-hour (Activity 24/10) or 5-hour (Activity 24/5) windows. It is assumed that the algorithms in the current study operate on similar principles, detecting oestrus based on deviations from a presumed stable baseline activity. However, this method is more effective when cow movement is confined, such as in barns or paddocks. In contrast, the varying grazing distances in this study determined by herdsmen likely disrupted the stability of the rolling mean, reducing the system’s sensitivity. This may explain the notably low CowAlert ^® sensitivity of 20% observed here, compared to sensitivities of up to 80% reported by Dolecheck et al. (2015) and 84% by McGowan et al. (2007).

The sensitivity of Estrotect™ heat detectors was 26%, closely aligning with that of CowAlert™. Given that the cows were primarily managed under grazing conditions, the Estrotect™ patches were more susceptible to environmental wear, which likely compromised their functionality. Prolonged exposure to elements such as mud, dung, and mechanical abrasion resulted in the patches being observed as dirty, torn, partially detached, or faded factors that may have interfered with accurate assessment of activation. Additionally, the adhesive site of the patches appeared to serve as a refuge for ticks, which prefer warm, moist body regions. Notably, the effectiveness of Estrotect™ varies across studies: Johnson and Jaeger (2016) reported a sensitivity of 79% in prepubertal heifers, while van den Berg (2014) found a sensitivity ranging from 34% to 56% when Estrotect ^® was used in combination with other oestrus detection methods.

Parity and stage of lactation (days in milk, DIM) emerged as the two key factors significantly influencing the performance of oestrus detection tools (ODTs). Parity, which reflects both the age and physiological maturity of the cow, has been associated with increased oestrus intensity. Previous studies have demonstrated that the number of mounts per hour rises from approximately 5.5 in nulliparous heifers to 7.9 in cows beyond their fourth parity, indicating that oestrus expression tends to intensify with successive lactations possibly due to behavioural conditioning from earlier oestrus cycles ( Gwazdauskas et al., 1983). Based on this, it would be reasonable to expect that the sensitivity of Estrotect™ would improve with advancing parity. However, findings from the present study contradict this assumption, as Estrotect™ sensitivity actually declined progressively with higher parity, showing the lowest mean sensitivity in cows of 5th to 9th parity. An alternative explanation considers the role of increased body weight and social dominance in older cows, which may disrupt the dynamics of sexually active groups. Supporting this view, earlier research has shown that over 60% of mounting behaviour is initiated by dominant cows in higher social hierarchies suggesting that such cows are more likely to initiate mounting rather than being mounted themselves.

For CowAlert ^®, the trend in sensitivity across parities was less consistent, although cows in the fourth parity exhibited the lowest mean sensitivity. Overall, a declining trend in mean sensitivity with increasing parity was observed, aligning with earlier findings that oestrus-related walking activity diminishes as the number of lactations increases, and that oestrus is more prominently expressed in cows of lower parity. A study conducted in France on Holstein cows similarly reported a significant parity effect, where first-parity cows had higher oestrus detection sensitivity using activity meters compared to those in their third parity ( Chanvallon et al., 2014). In contrast, P4 Rapid™ appeared minimally affected by parity, as its mean sensitivity remained relatively stable across different parities.

A statistically significant difference was observed across stages of lactation, with the most notable variation in sensitivity of the three oestrus detection tools occurring in cows sampled within the first 60 days in milk. In contrast, the French study reported an increase in oestrus detection sensitivity as lactation progressed.

Inadequate or failed oestrus detection, which delays conception, imposes a measurable financial burden. No clear studies showing the cost of missed oestrus events in this production system were found in the literature. Therefore, the study used the values estimated by Esslemont et al., 20001, for UK farms for average producing cows. Their model estimated cost of delayed conception ranges from £1.73 to £3.55 per cow per day for average-yielding cows failing to conceive between 85–100 days and 146–175 days postpartum, respectively. Since oestrus detection tools (ODTs) serve as screening tests, their results influence whether a cow is presented for artificial insemination (AI). In AI-based dairy systems, accurately identifying cows in oestrus is essential to optimize reproductive efficiency. Failure to detect oestrus means the cow is not inseminated during that cycle, incurring a loss of £56 per cow per missed 21-day cycle. Conversely, correctly identifying non-oestrus cows conserves labour and breeding resources, estimated at approximately £10 per instance.

Based on each ODT’s sensitivity, the cost of missed heats was simulated in a hypothetical 100-cow herd. Among the tools, P4 Rapid incurred the lowest cost (£150), while Estrotect™ resulted in the highest (£5,550). Similarly, false positives driven by lower specificity were simulated, with P4 Rapid showing the highest cost (£196) and Astrotech™ the lowest (£32).

Overall, the total combined cost of missed and false oestrus detections per cycle ranged from £150 to £5,550 per 100 cows. Thus, P4 Rapid proved to be the most economically advantageous per cycle per 100 cows compared to the other tools.

This study evaluated the effectiveness, robustness, and degree of agreement among three oestrus detection tools – P4 Rapid™, Estrotect™, and CowAlert ^® under tropical smallholder dairy conditions in Kenya. Using a progesterone-based reference standard, the study identified 130 oestrus events across 51 lactating cows and assessed each tool’s diagnostic performance and influencing intrinsic factors. P4 Rapid™ demonstrated superior sensitivity and predictive value for detecting true oestrus events, making it the most effective tool among the three. Estrotect™ and CowAlert ^® exhibited lower sensitivity but maintained high specificity, which may be beneficial in minimizing false positives. The performance of CowAlert ^® was significantly influenced by management system constraints, such as grazing distance and environmental variability, while Estrotect™’s durability was compromised by field exposure to dirt, moisture, and physical abrasion. P4 Rapid™’s performance remained relatively stable across different cow-level factors. Cow-level factors such as parity, days in milk, body condition, and locomotion scores significantly influenced tool sensitivity and specificity. Among these, parity and stage of lactation appeared most critical, particularly affecting Estrotect™ and CowAlert ^® performance. Economically, the use of P4 Rapid™ resulted in the lowest combined cost of missed and false oestrus events per cycle. Therefore, in resource-limited and grazing-based systems, P4 Rapid™ presents a more viable oestrus detection strategy despite its lower specificity. Overall, this study highlights the need to match oestrus detection technologies with farm-specific conditions and cow-level variables to improve reproductive efficiency and economic returns in smallholder dairy systems. Further research may explore integration or combination of multiple detection strategies to enhance accuracy under field conditions.