In December 2023, a systematic review was conducted by performing an online search of scientific studies in indexing databases, such as PubMed, Scopus, and Web of Science. The search utilized the PICo methodology, which focuses on Population, Interest, and Context of the research (Schardt et al., 2007). Thus, the search key used a combination of keywords to define population (e.g., broiler, chick, chicken, poultry), interest (e.g., phosphorus, P level, phosphorus level), and context (e.g., performance, body weight, average daily gain, weight gain, average daily feed intake, feed intake, feed consumption, feed conversion, feed to gain, feed:gain, feed efficiency, gain to feed, bone quality, bone characteristics, tibia weight, tibia ash, and shear force).

All articles found in each database were exported to the EndNote X9 software, which allowed the organization of the bibliographic references obtained in the indexing databases. The PRISMA flow diagram (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) was applied to select the articles (Moher et al., 2009). The title and abstract of each study were reviewed and evaluated according to the following selection criteria: 1) articles in English with experiments carried out in broiler chickens; 2) the presence of at least three different levels (treatments) of aP in the diets; and 3) reported growth performance results (weight gain, average daily intake, conversion, and feed efficiency) according to the diet phase and/or bone characteristics (tibia P, tibia ash, and shear force). In addition, as a complement to the first search, bibliographic references listed in the selected articles were also reviewed to search for additional studies that could meet the selection criteria for inclusion in the databases. The full version of the selected studies was critically evaluated for their quality and relevance, considering the objective of this systematic review.

Studies in the database were published between 1997 and 2023. Additionally, as a complement to the search, all reference lists of the selected studies were reviewed to search for additional studies that met the selection criteria for inclusion in the database. These scientific publications were evaluated according to the same selection criteria used in the previous step. After the final selection, data from the articles were transferred to a Microsoft Excel^® spreadsheet, where each row of the spreadsheet represented a treatment and each column represented a variable. For each article/treatment, two moderator codes were applied: a) general code (study effect): each dose-response study received a sequential number; and (b) intra-code: composed of the general code plus a sequential number representing the treatment (e.g., article 1, treatment 01 – 1 + 01 = 101). Growth performance results, such as average daily weight gain (ADG), average daily feed intake (ADFI), feed conversion ratio (FCR), and feed efficiency (FE), were collected as raw data and later used as relative information.

To reduce heterogeneity among studies, performance responses from each treatment were expressed relative to the response observed in the treatment with the highest value within the same experiment. This standardization approach allowed comparisons across trials despite differences in experimental design and response magnitude. For instance, if in study A, the highest ADG was 55 g/day and in study B it was 52 g/day, these values were normalized to 100% within each study, and all other treatments were scaled proportionally (e.g., the treatment with the highest ADG in each study was considered 100%). Although this method may obscure absolute performance differences, it preserves within-study treatment effects and facilitates meta-analytical modeling.

The consumption of aP was also standardized relative to the BT recommendations (i.e., the values were considered 100% if they were totally following the recommendations in the tables). Models available in the most recent version of the BT (2024) were used, as follows:

Equation I - 1 to 21 days: y = ( 0.026 × B W 0.75 ) + ( 6.0 × A D G )

Equation II - 22 to 42 days: y = ( 0.026 × B W 0.75 ) + ( 5.0 × A D G )

in which y is the aP requirement, in g of aP per day; BW is the body weight, in kg; and ADG is the daily weight gain, in kg.

The BT was selected as a baseline due to its updated prediction models and relevance in the Brazilian market. This allowed a consistent comparison of relative aP intake across diverse studies, while recognizing that absolute values may vary depending on local practices.

The database was divided into three phases: starter (1 to 10 d), grower (11 to 21 d), and finisher (≥ 22 d). The studies were assigned to each phase, based on the initial and final ages of the experimental birds. Studies that included ages belonging to two of the phases described above were assigned to the one in which most of the study period belonged (i.e., if a study was conducted with broilers from 1 to 15 days, that study was considered to be in the starter phase (10 days) and not in the grower phase (only 5 days)).

Characteristics such as bibliographic information, genetics, and sex were used to perform a descriptive and graphical analysis of the studies present in the database to verify the coherence of the data. In the modeling stage, aP intake was considered the independent variable in a series of sequential analyses performed individually for each dependent variable (growth performance or bone mineralization characteristics). Only studies with at least one level lower and at least one level higher than the recommended aP estimated in BT were considered for modeling. In addition, treatments with aP content higher than 140% of the recommendations of the test model were also removed. This cutoff was applied because excessive dietary aP content can form insoluble complexes in the intestinal lumen, which reduce nutrient digestibility and impair performance. Therefore, data points with aP intake far above practical feeding levels were considered physiologically unrealistic and potentially confounding for model fitting.

Exponential regressions were first adjusted by the NLIN procedure using subgroups of data (one equation for each article). The data obtained were explored using descriptive statistics and were used as starting points in the following procedure. At this stage, new models were adjusted using the NLMIXED procedure to combine the models generated by each dose-response treatment present in the database, considering the random effect of the studies included. This method accounts for unmeasured inter-study variability and reduces the risk of confounding. Thus, a general model was obtained for each dependent variable that considered the variability among the studies. Finally, a new analysis was performed using the ‘by’ option to generate specific equations for each production phase (starter, grower, and finisher). The nonlinear common plateau asymptotic regression equation fitted was:

Later, the optimal level of 𝑥 was obtained by the following equation:

in which 𝑦 is the dependent variable; 𝑥 is the relation between intake and current requirement of aP; 𝑏0 is the response for the dependent variable estimated for the feed with the lowest level of 𝑥; 𝑏1 is the difference estimated between the minimum and maximum response obtained by increasing 𝑥; 𝑏2 is the slope of the exponential curve; 𝑏3 is the 𝑥 at the lowest level. All analyses were performed using SAS statistical software (SAS Institute, v. 9.3).

The PRISMA diagram illustrating the studies identified and selected at each stage of the literature search is presented in Figure 1. After searching the indexing databases, 1,934 references were identified and imported from the databases from which 748 duplicate references were removed. In the exclusion stage, based on the titles and abstracts of the studies, 1,002 studies were excluded because they did not fall within the established scope. Furthermore, 94 studies were excluded because they included fewer than three levels of aP, did not report performance responses, or reported responses inconsistent with the diet phase. After evaluating the full articles and their references, 90 studies were selected and included in this meta-analysis.

Figure 1

The studies included in the database were published between 1997 and 2023 (Table 1). Only 3 studies did not report aP levels, and 43% of the studies reported phytase treatments. Most studies were conducted in the United States and Brazil, with 19 and 16 articles, respectively (Figure 2). The selected studies included 74,519 broilers, of which 6% were mixed groups (females and males), 71% were male, 8% were female, and 15% were not described. A wide diversity of genetic lineages has been described in the articles. Of these, the most cited were Ross and Cobb.

Table 1

Code	Reference		aP1 (%)	Variables analyzed2
				ADG	ADFI	FCR	FE	TA	TP	SF
1	Abdulla et al. (2016)		0.35-0.68	+	+	+	+	-	-	-
2	Abdulla et al. (2017)		0.35-0.68	+	+	+	+	+	+	-
3	Abudabos (2012a)		0.14-0.29	+	+	+	+	+	+	-
4	Abudabos (2012b)		0.13-0.40	+	+	+	+	-	-	-
5	Abudabos (2012c)		0.13-0.40	+	+	+	+	+	-	-
6	Adeola and Sands (2004)		0.40-0.55	+	+	+	+	+	-	-
7	Adeola and Walk (2013)		0.12-0.34	+	+	+	+	+	-	-
8	Aderibigbe et al. (2022)		0.12-0.44	+	+	+	+	-	-	-
9	Ahmad et al. (2023)		0.25-0.45	+	+	+	+	-	-	-
10	Akter et al. (2016)		0.30-0.40	+	+	+	+	+	+	-
11	Baradaran et al. (2014)		0.15-0.40	+	+	+	+	-	-	-
12	Bertechini et al. (2022)		0.20-0.40	+	+	+	+	+	+	-
13	Beyranvand et al. (2018)		0.39-0.49	+	+	+	+	+	+	+
14	Cabahug et al. (1999)		0.23-0.45	+	+	+	+	-	-	-
15	Cardoso Júnior et al. (2010)		0.27-0.42	+	+	+	+	+	-	-
16	Catalá-Gregori et al. (2007)		0.16-0.50	+	+	+	+	-	-	-
17	Cavalcanti and Behnke (2004)		0.35-0.50	+	+	+	+	-	-	-
18	Coon et al. (2007)		0.14-0.85	+	+	+	+	-	-	-
19	Coto et al. (2008a)		0.35-0.50	+	+	+	+	-	-	-
20	Coto et al. (2008b)		0.35-0.50	+	+	+	+	-	-	-
21	Cowieson et al. (2020)		0.35-0.48	+	+	+	+	-	-	-
22	David et al. (2022)		-	+	+	+	+	+	+	-
23	David et al. (2023)		-	+	+	+	+	+	+	-
24	Dhandu and Angel (2003)		0.10-0.31	+	+	+	+	+	-	+
25	Driver et al. (2005)		0.20-0.50	+	+	+	+	+	-	-
26	El-Sherbiny et al. (2010)		0.13-0.40	+	+	+	+	+	+	-
27	Garcia et al. (2006)		0.22-0.40	+	+	+	+	+	-	-
28	Gautier et al. (2017)		0.20-0.80	+	+	+	+	+	-	+
29	Gautier et al. (2018)		0.29-0.50	+	+	+	+	-	+	-
30	Gomes et al. (2004)		0.15-0.60	+	+	+	+	+	+	+
31	Gürbüz et al. (2009)		0.13-0.49	+	+	+	+	+	-	-
32	Hamdi et al. (2015a)		0.25-0.45	+	+	+	+	+	-	-
33	Hamdi et al. (2015b)		0.30-0.45	+	+	+	+	+	-	-
34	Hamdi et al. (2017)		0.27-0.42	+	+	+	+	+	-	-
35	Han et al. (2015)		0.20-0.45	+	+	+	+	+	+	+
36	Han et al. (2016)		0.35	+	+	+	+	+	+	+
37	Han et al. (2018)		0.25-0.65	+	+	+	+	+	+	-
38	Hassan et al. (2016)		0.21-0.45	+	+	+	+	-	-	-
39	Hu et al. (2020)		0.32	+	+	+	+	-	-	-
40	Imari et al. (2020)		0.28-0.48	+	+	+	+	+	+	-
41	Iqbal et al. (2023)		0.20-0.50	+	+	+	+	-	-	-
42	Iyayi et al. (2013)		0.14-0.23	+	+	+	+	+	-	-
43	Jiang et al. (2013)		0.15-0.45	+	+	+	+	+	+	+
44	Kahindi et al. (2017)		0.25-0.45	+	+	+	+	+	-	-
45	Karimi (2005)		0.29-0.45	+	+	+	+	-	-	-
46	Karimi et al. (2013)		0.15-0.45	+	+	+	+	+	-	-
47	Kiani and Taheri (2022)		-	+	+	+	+	-	-	-
48	Kwon et al. (2022)		0.10-0.40	+	+	+	+	+	-	-
49	Laurentiz et al. (2009)		0.17-0.45	+	+	+	+	-	-	-
50	Lee et al. (2017)		0.22-0.60	+	+	+	+	+	+	+
51	Leytem et al. (2008)		0.26-0.51	+	+	+	+	-	-	-
52	Li et al. (2016a)		0.27-1.13	+	+	+	+	-	-	-
53	Li et al. (2000)		0.20-0.45	+	+	+	+	+	-	-
54	Lim et al. (2001)		0.15-0.45	+	+	+	+	-	-	-
55	Lima et al. (1997)		0.15-0.45	+	+	+	+	-	-	-
56	Liu et al. (2016)		0.18-0.57	+	+	+	+	+	+	+
57	Mello et al. (2012a)		0.17-0.55	+	+	+	+	+	+	-
58	Mello et al. (2012b)		0.17-0.55	+	+	+	+	+	+	-
59	Miao et al. (2017)		0.23-0.68	+	+	+	+	-	-	-
60	Moghadam (2006)		0.20-0.30	+	+	+	+	+	+	-
61	Moradi et al. (2023)		0.33-0.48	+	+	+	+	-	-	-
62	Namini et al. (2012)		0.14-0.33	+	+	+	+	-	-	-
63	Nardelli et al. (2018)		0.11-0.34	+	+	+	+	+	+	-
64	Onyango et al. (2003)		0.13-0.50	+	+	+	+	+	-	+
65	Panda et al. (2007)		0.30-0.45	+	+	+	+	+	-	+
66	Peng et al. (2003)		0.29-0.45	+	+	+	+	-	-	-
67	Pereira and Adeola (2016)		0.12-0.35	+	+	+	+	-	-	-
68	Persia and Saylor (2006)		0.13-0.53	+	+	+	+	+	-	-
69	Powell et al. (2011)		0.40	+	+	+	+	-	-	-
70	Ribeiro et al. (2003)		0.16-0.36	+	+	+	+	+	-	+
71	Ribeiro Jr. et al. (2016)		0.18-0.45	+	+	+	+	+	+	-
72	Runho et al. (2001)		0.15-0.65	+	+	+	+	+	+	+
73	Santos et al. (2011)		0.29-0.47	+	+	+	+	+	-	-
74	Santos et al. (2019)		0.18-0.45	+	+	+	+	+	-	-
75	Sharma et al. (2018)		0.30-0.50	+	+	+	+	-	-	-
76	Shaw et al. (2011)		0.22-0.38	+	+	+	+	+	-	+
77	Silva et al. (2006)		0.25-0.45	+	+	+	+	-	-	-
78	Solomon et al. (2022)		0.23-0.43	+	+	+	+	+	+	-
79	Sun et al. (2018)		0.28-0.36	+	+	+	+	-	-	-
80	Tay-Zar et al. (2019)		0.23-0.53	+	+	+	+	+	-	+
81	Um et al. (2000)		0.12-0.45	+	+	+	+	-	-	-
82	Valable et al. (2018)		0.26-0.40	+	+	+	+	-	-	-
83	Vieira et al. (2015)		0.14-0.42	+	+	+	+	+	+	-
84	Viveros et al. (2002)		0.14-0.45	+	+	+	+	-	-	-
85	Waldroup et al. (2000)		0.10-0.50	+	+	+	+	+	-	-
86	Wang and Kim (2021)		0.20-0.44	+	+	+	+	+	-	-
87	Wilkinson et al. (2014)		0.25-0.55	+	+	+	+	+	-	-
88	Wu et al. (2004)		0.20-0.48	+	+	+	+	-	-	-
89	Yan et al. (2001)		0.10-0.45	+	+	+	+	+	-	-
90	Yan et al. (2004)		0.20-0.50	+	+	+	+	+	-	-

ADG - average daily gain; ADFI - average daily feed intake; FCR - feed conversion ratio; TA - tibia ash; TP - tibia phosphorus; SF - shear force.

Figure 2 1 In cases of international collaboration, the country where the experimental work was conducted was prioritized. When this information was not explicitly stated, the country of the first author's institutional affiliation was used.

In addition to age, the aP level recommended to optimize broiler growth performance varied depending on the variables evaluated. Using the exponential model, the relative P intake required to maximize ADG was 99% of the baseline BT recommendations, suggesting a slight trend toward overestimation (Figure 3). On the other hand, the grower (102%) and finisher (102%) phases showed a higher estimate of relative P intake, indicating a marginal potential underestimation. A slight overestimation of BT was identified for FE because the relative intake of P was 99% of the recommendations for the starter and grower phases (Figure 4). However, in the finisher phase, the estimate of relative P consumption reached exactly (100%) the recommendation suggested by the BT. Given the proximity of all these values to 100%, the model results should be interpreted as indicative of general alignment with BT values, rather than evidence for precise dietary adjustment.

Figure 3 1 Daily gain was expressed relative to the highest response in each study. Phosphorus intake was standardized to the Brazilian Tables for Poultry and Swine.

Figure 4 1 Feed efficiency was expressed relative to the highest response in each study. Phosphorus intake was standardized to the Brazilian Tables for Poultry and Swine.

The relative intake of P that maximized the bone mineralization characteristics of the broilers was higher than the recommendations established in the BT in all responses and in the three phases studied, suggesting an underestimation, especially in the final phase. The values obtained for maximum P deposition in the tibia ranged from 105% (starter) to 109% (grower and finisher) of the BT recommendations (Figure 5). A similar behavior was observed for the bone ash characteristics, with adjustment values between 101 (starter) and 111% (finisher) of the recommendation of the proposed models in BT (Figure 6). Finally, the shear force was maximized at levels between 101 (starter) and 112% (finisher) of the current recommendations of the tested models (Figure 7).

Figure 5 1 Tibia phosphorus was expressed relative to the highest response in each study. Phosphorus intake was standardized to the Brazilian Tables for Poultry and Swine.

Figure 6 1 Tibia ash was expressed relative to the highest response in each study. Phosphorus intake was standardized to the Brazilian Tables for Poultry and Swine.

Figure 7 1 Shear force was expressed relative to the highest response in each study. Phosphorus intake was standardized to the Brazilian Tables for Poultry and Swine.