Symposium review: Use of multiple biological, management, and performance data for the design of targeted reproductive management strategies for dairy cows*

As the reproductive efficiency of dairy cattle continues to improve in response to better management and use of technology, novel reproductive management approaches will be required to improve herd performance, profitability, and sustainability. A potential approach currently being explored is targeted reproductive management. This approach consists of identifying cows with different reproductive and performance potential using multiple traditional and novel sources of biological, management, and performance data. Once subgroups of cows that share biological and performance features are identified, reproductive management strategies specifically designed to optimize cow performance, herd profitability, or alternative outcomes of interest are implemented on different subgroups of cows. Tailoring reproductive management to subgroups of cows is expected to generate greater gains in outcomes of interest than if the whole herd is under similar management. Major steps in the development and implementation of targeted reproductive management programs for dairy cattle include identification and validation of robust predictors of reproductive outcomes and cow performance, and the development and on-farm evaluation of reproductive management strategies for optimizing outcomes of interest for subgroups of cows. Predictors of cow performance currently explored for use in targeted management include genomic predictions; behavioral, physiological, and performance parameters monitored by sensor technologies; and individual cow and herd performance records. Once the most valuable predictive sources of variation are identified and their effects quantified, novel analytic methods (e.g., machine learning) for prediction will likely be required. These tools must identify groups of cows for targeted management in real time and with no human input. Despite some encouraging research evidence supporting the development of targeted reproductive management strategies, extensive work is required before widespread implementation by commercial farms.


INTRODUCTION
As the dairy industry evolves and new strategies are required to improve herd performance, management, and sustainability measures, a potential approach to improve reproductive efficiency, optimize herd management practices, and increase profitability is through targeted reproductive management (TRM).Also known as "precision" or "personalized" management (Berry et al., 2016;Zolini et al., 2019), this approach consists of identifying and characterizing predictors of cow reproductive and performance outcomes using multiple traditional and novel sources of biological, management, and performance data.Once identified, subgroups of cows with unique biological features or expected performance are targeted with management strategies specifically designed to optimize cow performance, herd profitability, or alternative outcomes of interest (Figure 1).The expectation is that by tailoring reproductive management to subgroups of cows, greater gains in outcomes of interest will be realized than if the whole herd is under similar management.Examples of TRM include submission to AI with programs that prioritize AI at detected estrus (AIE) or timed AI (TAI; Fricke et al., 2014;Giordano et al., 2015;Pérez et al., 2020b), increasing the economic value of offspring and reducing herd replacement costs through targeted use of sexed semen, beef semen, valuable high genetic merit bull semen, or embryo transfer (Kaniyamattam et al., 2018;Berry, 2021); targeted hormonal therapy (Bisinotto et al., 2015;Giordano et al., 2015;Zolini et al., 2019); manipulation of timing of pregnancy during lactation (Gobikrushanth et al., 2014;Stangaferro et al., 2018a,b); and insemination decisions based on Symposium review: Use of multiple biological, management, and performance data for the design of targeted reproductive management strategies for dairy cows* expected probability of success or value of pregnancy (Figure 1).As dairy farms adopt and make better use of novel data-driven technologies, adapt to increasing labor constraints, and change management practices in response to market conditions and consumer-driven trends, the suite of TRM strategies available and the motivations to use them will expand and evolve.
Major steps in the development and implementation of TRM programs for dairy cattle include identification and validation of robust predictors of reproductive outcomes and cow performance, and the development and on-farm evaluation of reproductive management strategies for optimizing outcomes of interest for subgroups of cows.Therefore, the purpose of this review is to discuss the rationale behind the development and implementation of TRM strategies for dairy cattle.Special emphasis is placed on recent research aimed at identifying and characterizing predictors for use in TRM, and on-farm evaluation of TRM.Finally, current challenges, research knowledge gaps, and future opportunities are discussed.

PREDICTORS FOR USE IN TRM
A critical step in the development of TRM strategies is the identification and characterization of predictors of variability in reproductive outcomes of interest for individual or subgroups of cows.These predictors might be the cause of variation or simply associations between biological, management, or environmental parameters with reproductive performance and other relevant dairy herd outcomes.In isolation or through multiple complex interactions, these predictors explain part of the observed variation in reproductive performance among dairy cattle.Many associations between predictors and reproductive outcomes are well-known and characterized, whereas others are yet to be discovered or are currently poorly characterized because of limitations to generate and evaluate data for large number of cows under commercial farm conditions.Until recently, data capture to characterize these predictors was either too cumbersome or expensive.In contrast, continued adoption of technologies for data collection and herd management in commercial farms enables collection of comprehensive data sets from large numbers of cows with less difficulty and at lower cost.Currently available and emerging technologies offer unprecedented opportunities to capture novel data in large amounts, fully or semi-automated, and in some cases in real time (Bruinjé et al., 2019;Halachmi et al., 2019;Schilkowsky et al., 2021).In parallel, tools and resources for seamless data integration and real-time data analytics for decision-support are becoming available, facilitating innovative and efficient use of data (Cabrera et al., 2020;Pérez et al., 2020aPérez et al., , 2021)).
The number and variety of predictors with consistent associations with reproductive outcomes is substantial.Therefore, we will focus on a few of the known predictors and sources of variation of reproductive outcomes that hold promise for use in TRM and for which data from large numbers of cows in commercial farms has become more accessible.

Sensor Data
Existing and emerging sensor-based technologies for monitoring individual cow behavioral (e.g., physical activity, rumination time, eating behavior, resting time), physiological (e.g., body temperature, milk conductivity, body condition score, rumen pH), and performance (e.g., milk yield and components) parameters in real time provide unprecedented opportunities to generate predictive tools for use in TRM.As these technologies become more widespread in commercial dairy farms (Drewry et al., 2019), data can be generated from more cows, at more frequent intervals, noninvasively, without cow manipulation, and at a lower cost than with other methods.Therefore, it is vital to improve our understanding of the associations among sensor-generated parameters and reproductive outcomes.Similarly, the influence of management practices and environmental conditions on sensor-generated predictors must be determined and quantified because management and environmental conditions can dramatically affect patterns of sensor-generated parameters and their associations with predictors and outcomes of interest.Data generated by sensors included in automated monitoring and management systems might be used in different manners to create subgroups of cows for TRM.Examples include (1) records of occurrence of discrete reproductive biological events such as estrus, ovulation (or lack thereof), and pregnancy; (2) features from discrete events recorded by sensors such as estrus intensity and duration; and (3) sensor parameter pattern features associated with reproductive outcomes.
An example of the first type of sensor data use consists of using estrus event alerts generated by automated estrus detection (AED) systems during the voluntary waiting period (VWP).Once subgroups of cows with different reproductive potential (i.e., probability to express estrus, probability of pregnancy at first service) based on expression of estrus are identified, cows can be assigned to different TRM strategies.This approach is supported by recent data from multiple studies, such as a retrospective study that grouped lactating cows from 5 commercial farms in Germany based on the number of estrus events detected by an AED system from 7 to 40 DIM (Borchardt et al., 2021).Cows with at least 1 or 2 estrus events recorded had greater odds of insemination before 100 DIM, received first AI earlier, and had a greater hazard of pregnancy up to 200 DIM.Moreover, compared with cows with no recorded estrus events in early lactation, those with at least 1 estrus alert had a more intense estrus event at the first postpartum AI, which is associated with greater probability of pregnancy (Madureira et al., 2015;Burnett et al., 2018;Madureira et al., 2019).In a recent experiment from our group designed to evaluate TRM strategies (Rial et al., 2021), we observed that cows with at least 1 automated estrus alert from 21 to 50 DIM and submitted to programs that prioritized AIE were more likely to be inseminated at detected estrus (87.2 vs. 51.6% for cows with or without automated alerts).Similarly, cows with at least 1 estrus alert during the VWP were more likely to become pregnant at first service (46.0 vs. 37.9% P/ AI for cows with or without automated alerts), regardless of submission to first AI at detected estrus or TAI after a Double-Ovsynch protocol.Evidence from these studies supports the use of automated estrus alerts during the VWP as a tool to identify cows with different likelihood of estrous expression and fertility potential after the end of the VWP.
An example of the second type of sensor data use in TRM, namely using specific features from sensorrecorded events, is the use of estrus alerts duration and intensity as predictors of reproductive outcomes.In this regard, data from multiple studies demonstrated an association between intensity of estrus events as determined by AED systems and the probability of pregnancy after AI (Madureira et al., 2015;Burnett et al., 2018;Madureira et al., 2019).Cows with greater physical activity levels and reduced lying time during estrus had greater P/AI than cows with lesser intensity of estrus.Cows with more intense estrus had improved P/AI with differences in the range of 10 to 14 percentage points as compared with cows with less intense estrus.Another recent retrospective study from Germany using data from 5,933 AI services from 3,132 cows at 8 commercial dairy farms reported that cows with high estrous intensity based on physical activity levels had 1.35 greater odds of pregnancy after AI than cows with low intensity of estrous (Tippenhauer et al., 2021).Collectively, data from these studies suggested that combining estrus alert records and specific estrus event features generated by AED systems (i.e., intensity of estrus) during and after the VWP, could be a useful tool for creating of subgroups of cows for TRM.
Associations between sensor-monitored parameters during estrus and reproductive outcomes are likely the strongest as they are observed closer to AI. Nev-ertheless, it is also plausible that certain features of sensor-generated parameters recorded for weeks or months before AI could be associated with reproductive outcomes.Of particular interest are parameters indicative of the health and metabolic status of cows in late gestation and early lactation because of the wellknown associations between early-lactation cow health and metabolic status with reproductive performance (Ospina et al., 2010;Chapinal et al., 2012;Manríquez et al., 2021).Understanding and characterizing associations from these early predictors would enable decisionmaking sometime before cows become eligible for pregnancy.For example, a recent study demonstrated an association between rumination time in the week before calving and pregnancy risk (Cook et al., 2021).As the weekly average for rumination time 1 wk before calving increased by 15 min, the risk of pregnancy increased by 4% (95% CI: 1-7%).In a recent study from our group (Granados et al., 2020), we compared the pattern of multiple sensor-generated parameters before firstservice TAI for lactating cows that became pregnant or not after first service.We observed that multiple parameters recorded from ~4 wk before calving until 56 DIM were different for pregnant and nonpregnant cows after first AI.For example, we observed differences for number of resting bouts and milk fat and protein yield for primiparous cows and fat percent, fat-to-protein ratio, rumination time, physical activity, reticulo-ruminal temperature, and BW for multiparous cows.As the differences observed were parity specific, our data suggested that the type of parameters and prediction tools used may need to be customized for cows of different parity.Indeed, customization of algorithms to adjust for variation in sensor parameters might be relevant to adjust not only for variation among cows or subgroups of cows that share certain biological features, but also across farms, given the potential substantial effects of farm-specific conditions on the behavior of sensormonitored parameters.
Ultimately, as the direction and strength of associations between cow behavioral, physiological, and performance parameters collected by automated sensor technologies and reproductive outcomes of interest are better understood, sensor data could be used alone or in combination with other nonsensor data for improving predictions of cow reproductive performance for use in TRM.

Genomic Predictions of Reproductive Performance
A relationship between genetics and reproductive performance in dairy cattle has long been recognized and characterized.More recently, the availability of genomically enhanced predictions for fertility for dairy females enabled the evaluation of reproductive performance and physiological outcomes in cows of different genetic merit for fertility in more detail under controlled conditions in commercial dairy farms.For example, Lima et al. (2020) reported multiple associations between genomic daughter pregnancy rate (GDPR) and reproductive outcomes for primiparous and multiparous lactating Holstein cows in commercial dairy farms submitted to AI with a program that combined AIE and TAI through synchronization of estrus and ovulation with a Presynch-Ovsynch protocol.Primiparous and multiparous cows in quartiles of greater GDPR for the population studied had fewer days to first AI, greater P/AI at first service, fewer services per conception, fewer days to pregnancy, and more cows were pregnant at the end of lactation.For primiparous cows, the proportion of cows AIE increased for cows in quartiles of greater GDPR, suggesting greater ability to express estrus for cows of greater genetic merit for fertility.Similarly, using data from primiparous Holstein cows from a commercial dairy farm, Chebel and Veronese (2020) reported beneficial associations between GDPR and reproductive performance outcomes.Cows with greater genetic merit for fertility based on GDPR had greater hazard of AIE, number of estrus events before 62 DIM, and fewer days to pregnancy.In a recent study from our group (Sitko and Giordano, unpublished data, Cornell University), in which 2,534 primiparous cows from 6 commercial herds in New York were assigned to reproductive management programs that prioritized AIE or TAI with fertility programs, we observed that cows in the highest GDPR quartile had improved reproductive performance compared to cows in the lowest GDPR quartile.Cows of superior genetic merit for fertility had greater P/AI to first service regardless of the method of submission to AI (Table 1), became pregnant earlier (P < 0.05) during lactation [hazard ratio (HR): 1.36; 95% CI: 1.20-1.53],and there was a tendency (P = 0.06) for more pregnant cows at 200 DIM (91.1 vs. 86.9%for highest vs. lowest GDPR quartile).In addition, a smaller proportion of cows in the lowest (65.5%) than the highest (74.5%) quartile for GDRP received AIE when managed with the program that prioritized detection of estrus rather than TAI.For most outcomes, cows in the medium quartiles had intermediate performance in line with genomic predictions.These multiple studies demonstrated consistent associations between reproductive performance and genomic predictions across a wide range of commercial farm management conditions, and the dissimilar physiological milieu and insemination dynamic resulting from programs that prioritize AIE or TAI.Thus, Giordano et al.: REPRODUCTION SYMPOSIUM genomic predictions used alone or in combination with other predictors might be a suitable tool for identifying subgroups of cows for TRM.

Other Outcome Predictors for Use in TRM
Although genomic predictions for fertility and sensor-generated data are promising candidates for generation of subgroups of cows for TRM, many other predictors of reproductive outcomes and factors known to consistently affect reproductive performance could be used in the design and implementation of TRM strategies.Cow biological features and factors such as parity, health status and occurrence of health events in early lactation, calving-related events, cyclicity status, BCS (Manríquez et al., 2021), and milk production level have all been associated with reproductive outcomes (Chapinal et al., 2012;Pinedo et al., 2020).For example, parity number is a simple to use cow factor strongly associated with reproductive outcomes such as probability of expression of estrus and fertility (Madureira et al., 2015;Pinedo et al., 2020).On the other hand, management and environmental conditions are known to consistently influence reproductive biology and performance of cows (De Rensis and Scaramuzzi, 2003;Schefers et al., 2010) to an extent that variation in these factors could be used to identify subgroups of cows for TRM.Unlike sensor-generated parameters and genomic predictions, several of the cow, management, and environmental factors suitable for use in TRM are routinely collected or easily recorded at no extra cost.
In summary, there is ample evidence of numerous consistent associations between predictors and outcomes of interest.Moreover, differences among cows across levels of continuous or categorical predictors were of sufficient magnitude to enable grouping cows such that differences among groups for outcomes of interest justify TRM.As no single predictor explains all the variation in a cow population, and multiple factors might affect associations between predictors and outcomes, combining multiple predictors might be beneficial for creating subgroups of cows for TRM.

DEVELOPMENT AND ON-FARM EVALUATION OF TRM STRATEGIES
Because of the many reproductive management decision-making steps in the lifespan of dairy cattle, the diversity of reproductive tools available, and the variety of outcomes of interest for optimization or improvement, there are numerous options for the development of TRM programs.In recent years, experiments conducted on commercial farms explored different forms of TRM including programs designed based on automated estrus alerts during the VWP or around insemination, genomic predictions for fertility and cow genotype, ovarian physiological status at nonpregnancy diagnosis, and others.

Automated Estrus Alerts
Based on the known association between estrus expression during and after the VWP (Borchardt et al., 2021), first-service management programs that combine a period of AIE followed by TAI have been tailored for cows with (i.e., estrus cows) or without estrus (i.e., noestrus cows) events recorded by AED systems during the VWP.Fricke et al. (2014) evaluated herd effects of this type of TRM program.Cows with at least 1 AED system alert during a 53-d VWP were allowed up to 26 d of AIE, whereas no-estrus cows were allowed only 12 d for AIE based on AED system alerts.Despite a reduction in overall first-service P/AI for the TRM program, similar pregnancy rate and proportion of cows pregnant during lactation were observed when compared with an all-TAI program using Presynch-Ovsynch for ovulation synchronization and longer VWP (i.e., 79 d).Thus, the TRM program was effective at reducing the number of hormonal interventions for synchronization of estrus and ovulation without negatively affecting overall herd reproductive performance as measured by pregnancy rate.In addition, hormonal intervention was targeted at cows that otherwise would have had extended DIM at first service if they did not receive TAI.This is an example of using TRM not necessarily to increase reproductive performance but rather as an approach to optimize aspects of dairy herd management such as reducing hormonal interventions for TAI or maximizing the use of an AED system.Reducing hormonal intervention without compromising reproductive performance might become more relevant for commercial farms as scrutiny and negative perception of technology use in livestock farming increases (Pieper et al., 2016).An important consideration for the development of TRM programs based on probability of expression of estrus after the VWP is duration of the period of AIE.
In a recent experiment from our group (Rial et al., 2021), we observed that reducing the duration of the AIE period from 21 to 14 d for no-estrus cows during the VWP reduced days to first service by ~3 d without a reduction in the proportion of cows AIE.Conversely, extending the AIE period from 21 to 28 d for estrus cows did not increase the proportion of cows AIE and extended DIM to first service by ~1.5 d.These data suggested that providing an AIE period of ~2 wk for no-estrus cows during the VWP would allow enough time for most cows expected to express estrus to receive AIE while hastening insemination of cows more likely to need TAI.Conversely, for estrus cows during the VWP, extending the AIE period beyond 21 d might not necessarily increase the proportion AIE and could delay first insemination.
Recently, it has also been demonstrated that automated estrus alert features could be used to identify subgroups of cows that benefit from targeted hormonal intervention.For example, Burnett et al. (2022) reported that treatment with GnRH at the time of insemination at detected estrus with AED systems increased P/ AI of lactating cows by ~6 percentage points in cows with lesser intensity of estrus based on the median intensity value for individual farms.Conversely, cows considered to have greater intensity of estrus did not benefit from GnRH treatment.Implementing this or similar hormonal interventions at the time of insemination could become widespread given its simplicity and low cost for commercial farms that already implement automated detection of estrus that generate data for intensity of estrus.In addition, targeting hormonal intervention can help increase overall herd fertility and minimize unnecessary use of hormones in cows that do not benefit from the treatment.

Cow Genotype and Genomic Predictions
Genomically enhanced predictions for fertility and cow genotype have also been proposed and explored as the basis for TRM.Zolini et al. (2019) reported that the fertility-enhancing effect of human chorionic gonadotropin (hCG) treatment post-AI depended on parity and, more interestingly, on variation in cow genotype.Cows treated with hCG 5 d after AI for increasing circulating progesterone (P4) levels during early embryo development had a different response to hCG depending on allele variations for the coenzyme Q9 (COQ9) gene.Cows with the AA allele for COQ9 that received hCG had decreased P/AI, cows with the AG allele had increased P/AI, whereas cows with the GG allele had no difference in P/AI when compared with cows treated with vehicle control.Therefore, it has been suggested that as the number of cows genotyped in commercial farms increases, genotype data can be used as a strategy to target hormonal therapy for enhancing fertility of cows expected to have the greatest positive response and avoid treatment of cows that might not benefit or for which the therapy may be detrimental.More recently, in a study from our group (Sitko and Giordano, unpublished data, Cornell University) in which 2,534 primiparous cows from 6 commercial herds in New York were assigned to a reproductive management program that prioritized AIE (i.e., first AI: AIE after a single PGF 2α for 21 d and then P4-Ovsynch for TAI and 2+ AI: P4-Ovsynch at 32 d after a previous AI) or a program that prioritized TAI with fertility programs [i.e., first AI: Double-Ovsynch and 2+ AI: D25-Resynch for corpus luteum (CL) cows and P4-Ovsynch for NoCL cows], we observed an interaction between genetic merit for fertility based on GDPR and the response to reproductive management.Interestingly, cows in the highest GDPR quartile had a greater hazard of pregnancy (HR: 1.34; 95% CI: 1.14-1.58)for up to 200 DIM after calving when managed with the program that prioritized AIE, whereas cows in the lowest GDPR quartile had no difference in pregnancy rate (HR: 1.07; 95% CI: 0.90-1.26).These data suggested that reproductive performance of cows of different genetic fertility potential might be optimized with different types of reproductive management strategies, and thereby ge-Giordano et al.: REPRODUCTION SYMPOSIUM nomic predictions could be used as a tool to generate subgroups of cows for TRM.

Ovarian Physiological Status
Targeted hormonal therapy for increasing fertility to TAI or the proportion of cows AIE has also been explored for cows with different ovarian physiological status at the beginning or during synchronization of ovulation.Bisinotto et al. (2015) demonstrated the benefit of targeted P4 supplementation using CIDR devices for cows without a CL at the initiation of the Ovsynch protocol used for submission of cows to first service.Cows without a CL (NoCL cows) detectable by transrectal ultrasonography that received 2 CIDR devices had greater fertility than untreated cows, and similar fertility to cows with a CL (CL cows) present on the ovaries at the beginning of Ovsynch.Similarly, targeted re-initiation of Ovsynch with P4 supplementation for NoCL cows at nonpregnancy diagnosis in cows enrolled in a Resynch program was shown effective to increase P/AI (Wijma et al., 2018).Targeted induction of estrus expression with PGF 2α for CL cows and Ovsynch with P4 supplementation for NoCL cows at nonpregnancy diagnosis was effective for increasing the proportion of cows AIE and the fertility of NoCL cows while achieving similar pregnancy rates than with blanket use of the Ovsynch protocol for resynchronization (Giordano et al., 2015).
These multiple experiments provided evidence that dairy herd performance and management outcomes can be optimized through TRM.However, the type of outcome improved, and the extent of the improvement varied substantially depending on the strategy evaluated.Therefore, future selection of TRM programs for individual farms will depend not only on the resources and tools available to generate subgroups of cows, but also on the outcomes of interest for optimization.

CURRENT CHALLENGES, RESEARCH GAPS, AND FUTURE OPPORTUNITIES
Although progress has been made toward the development and implementation of TRM for dairy herds, numerous challenges remain before the dairy industry can take full advantage of this type of management.More research is needed to better understand existing and emerging predictors of outcomes of interest, the suite of TRM programs must be expanded and integrated, and, finally, software or other management tools for seamless on-farm implementation of TRM are needed to facilitate widespread adoption by commercial farms.

Predictors for TRM
Ideally, predictors for use in TRM programs present consistent associations with outcomes of interest, and the magnitude of the variation among subgroups of cows created based on the predictor is sufficient to justify targeted management.Ultimately, these features will dictate the predictive value, and thereby the accuracy in which subgroups of cows for targeted management are created.Even though some of the known associations have been consistent, and the magnitude of the differences among cows based on predictor levels was meaningful, associations have been more inconsistent or of lesser magnitude for other predictors (Madureira et al., 2019;Chebel and Veronese, 2020;Lima et al., 2020;Schilkowsky et al., 2021).Another factor that adds complexity to the identification of predictors for TRM is that some associations might be herd-, region-, or production-system specific (e.g., high-producing confinement systems vs. lower production grazing systems) due to variation in farm management practices and environmental conditions as well as variation in data capture, processing, and analysis.Indeed, dairy farm has been one of the greatest sources of variation for prediction of AI success in dairy cows in studies involving multiple operations (Schefers et al., 2010;Shahinfar et al., 2014;Rutten et al., 2016).A working model to circumvent these issues might consist of the development of a common process to create tools for TRM that can be implemented across farms, but that can be refined based on data that reflects individual farm conditions.The accuracy of predictions for creation of subgroups of cows for TRM is also relevant.The need for highly accurate predictions; however, will vary depending on the decision to be made.Specifically, the degree to which misclassification of cows in subgroups for targeted management will be problematic will vary depending on the risk or cost associated with the decision.For example, deciding not to inseminate a cow detected in estrus because of low predicted probability of pregnancy based on low estrous intensity may be more problematic than deciding not to treat a cow after AI with hormonal therapy for improving fertility.Thus, further research is needed to develop mechanisms for balancing the potential benefit of TRM with the risk of causing detrimental effects on herd outcomes.

Expanding and Integrating the Suite of TRM Strategies
Research conducted thus far has resulted in the development of a limited and narrow scoped number of TRM programs.Gains in efficiencies and improvements in herd management have been demonstrated, but it is unlikely that the true potential of TRM has been realized.Although attempts have been made to combine more than one strategy to optimize outcomes of interest (Sitko et al., 2019), in most cases, a single change to management has been implemented (e.g., hCG therapy after AI).Therefore, a future approach to untap the potential of TRM could consist of integrating multiple strategies or changes to management that complement or synergize with TRM.For example, a herd could use one or a series of algorithms to implement a comprehensive program that uses multiple predictors to assign cows to methods of submission to first AI that prioritize AIE or TAI based on the likelihood of estrus expression after the end of the VWP, assigns the type of semen for AI based on the likelihood of insemination success and expected genetic gains (e.g., sexed semen for cows most likely to become pregnant and beef for the rest), and uses post-AI hormonal therapy for cows expected to have low fertility after AI.The greatest challenge for integrating multiple TRM strategies is the vast and diverse number of options available for experimentation.

Tools for On-Farm Implementation
Systematic and efficient implementation of TRM in commercial dairy farms will depend upon the availability of software or other management tools incorporating algorithms for generating automatically, in real time, and with the least amount of human input subgroups of cows for TRM.The same or complementary tools are needed for assigning and implementing the selected targeted strategies for the specific subgroups of cows created.Algorithms for subgroup creation could be based on simple logic or advanced analytic tools such as machine learning algorithms.The former could be based on simple associations between outcomes and well-defined cow features (e.g., parity, breed), discrete events (e.g., estrus, health disorders in early lactation), and categorization of cows based on cutoff values for simple continuous outcomes (e.g., GDPR quartiles).On the other hand, machine learning algorithms or other forms of advanced data analytic tools could incorporate a greater number of more complex predictors for optimizing one or more outcomes of interest at the same time.Of particular interest would be the possibility of including complex associations between features of the pattern of behavioral, physiological, and performance parameters monitored by sensors and complex genomic traits.This will require powerful engines for data integration, processing, and analysis, which are currently under development (Cabrera et al., 2020;Pérez et al., 2020aPérez et al., , 2021)).Ultimately, these engines and algorithms must be developed to serve the needs of specific farms and be adaptable as the motivations and needs of dairy farms to implement TRM might change overtime.

CONCLUSIONS
Targeted reproductive management is an approach aimed at optimizing reproductive, performance, or other herd management outcomes of practical value for dairy farms for subgroups of cows that share biological features or are expected to have similar responses to specific management practices.Tailoring reproductive management to subgroups of cows is expected to generate greater gains in herd performance, economic, or management outcomes of interest than if the whole herd is under similar management.Technologies for capturing individual cow, herd, and environmental data in large amounts, fully or semi-automated, and in real time facilitate characterization of numerous known and novel associations among potential predictors for use in TRM and relevant outcomes.Recent research has demonstrated differences of sufficient magnitude among cows across levels of predictors as to justify formation of subgroups of cows for targeted management.It is also likely that combining multiple predictors might be required for creating subgroups of cows because no single predictor explains all the variation in a cow population, and multiple factors might affect associations between predictors and outcomes.Although different forms of TRM have been shown effective at increasing efficiencies or optimizing herd management practices, the suite of TRM programs evaluated and available for on-farm implementation must be expanded.In addition, research must be conducted to explore the potential synergies of integrating multiple targeted management practices.Major efforts are still needed to develop tools and resources that facilitate implementation of TRM in commercial farms.

Figure 1 .
Figure 1.Conceptual framework for development and implementation of targeted reproductive management (TRM) strategies.Major steps in the development and implementation of TRM programs include identification and validation of robust predictors of reproductive outcomes and cow performance.Examples include behavioral, physiological, and performance parameters collected by sensor technologies, genomic predictions for fertility, and health event data.Based on levels of one or more of these predictors, subgroups of cows expected to have similar level of performance or outcomes are created.Subgroups of cows are then targeted with reproductive management strategies aimed at optimizing outcomes of practical value for farms.Greater gains in herd performance are expected from optimizing specific outcomes for subgroups of cows than by applying similar management to the whole herd.AIE = AI at detected estrus; TAI = timed AI; P/AI = pregnancies per AI.
The authors' work was supported by the USDA National Institute of Food and Agriculture (NIFA; Washington, D.C.), Animal Health Program Project # 2017-67015-26772, Hatch project NYC-2020-21-255, and Multistate project 1021189.Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of NIFA or the United States Department of Agriculture (USDA).The authors have not stated any conflicts of interest.
Giordano et al.: REPRODUCTION SYMPOSIUM

Table 1 .
Effect of fertility group based on quartiles of genomic daughter pregnancy rate on pregnancies per AI (P/AI) 32 d after AI for first service for primiparous lactating Holstein cows