Energy Consumption Forecasting in Public Nursing Homes Using Multivariable Regression Models

Highlights

What are the main findings?

• Predictive models for energy consumption, energy costs, and CO₂ emissions were developed.
• The variables that significantly influence energy consumption were determined.

What are the implications of the main findings?

• Models support energy management benchmarking and decision-making.
• The results facilitate enhanced energy management within nursing home facilities.

Abstract

Buildings represent 40% of the European Union’s energy consumption and 36% of its greenhouse gas emissions. Nursing homes are among the buildings that consume the most energy. The objective of this study was to make predictive models of Energy Consumption, Energy Costs, and CO₂ Emissions in nursing homes using different variables. To do this, data from 20 public nursing homes located in Extremadura (Spain) during the 2019–2023 period were analyzed. All the buildings were built or renovated between 1995 and 2009; the useful area and the number of residents were in the range of 1332–10,880 m² and 24–254 residents. A statistical analysis was performed using multivariable linear regression. During the research, equations that allow for the estimation of the annual Energy Consumption, Energy Costs and CO₂ Emissions of nursing homes, according to the useful area and number of residents, were found. The R_adj² was 0.9710, 0.9744 and 0.9742, respectively. The quality of the models obtained was contrasted using the mean absolute error (MAE), the relative error (RE) and the root mean square error (RMSE), together with the assessment of multicollinearity through the Variance Inflation Factor (VIF). The findings of this study may prove beneficial for stakeholders within the elder care sector.

1. Introduction

The European Climate Law has made legislation the objective set in the European Green Pact for Europe’s economy and society to achieve climate neutrality by 2050 [1]. All newly constructed buildings are required to be zero-emission starting in 2030 [2], while existing ones should become certified zero-emission by 2050 at the latest [3,4]. Buildings represent approximately 40% of the EU’s overall energy demand and contribute to 36% of its greenhouse gas emissions [5].

Approximately 85% of buildings in the EU were constructed before 2000, of which 75% are poorly energy-efficient [6]. Taking action to improve building energy efficiency is therefore crucial to saving energy, reducing energy bills for citizens and small companies, and achieving a zero-emission, fully decarbonized building stock by 2050 [7].

With advances in medical technology and improvements in living standards, population aging has gradually become a global issue [8]. In the European Union in 2020, 21% of the population was aged 65 and over, compared to 16% in 2001 [9]. In Spain, the population aged over 65 years will currently reach a maximum of 20.4% and is projected to increase to 30.5% by around 2055 [10].

A nursing home is a residence that provides continuous care and support, overseen by qualified personnel [11], to individuals who are no longer able to remain in their own homes due to increasing dependence in performing daily activities, complex medical needs, and heightened vulnerability [12]. In Spain, there are 5188 nursing homes, with 381,514 places available [13].

Nursing homes are groups of buildings that consume significant amounts of energy and must remain in continuous operation throughout the day and year-round [14], resulting in higher energy consumption compared to other tertiary-sector buildings, such as commercial or educational buildings [15]. Nursing homes are among the structures with the highest energy consumption, attributable to their perpetual occupancy, elevated indoor temperatures, and a range of energy-consuming activities not typically found in other tertiary buildings. These activities include laundry, cooking, heating, cooling, ventilation, lighting, and operating medical devices, among others [16]. It is very important to evaluate and predict the energy consumption of nursing homes, since they are buildings with very intensive energy use [17], and because in the future it is expected that there will be more and more buildings destined for this use, given the aging population [18].

Vergés et al. (2024) developed a model based on artificial neural networks (ANN) to evaluate the energy consumption of the HVAC system during the cooling season in 8 nursing homes and concluded that by adjusting the operational temperatures adaptively up to 23.4% of energy savings can be achieved [19]. These results are better than those obtained through linear regression models (up to 9.9% energy savings) developed by Vergés et al. (2023) with the same data set [20].

Along these lines, recent studies, such as that by Li et al. (2025), have combined in situ measurements and simulations to analyze the energy performance and flexibility of buildings under different control strategies, including variations in operating temperatures, to develop more robust regression models [21]. In addition, system-optimization approaches, such as those proposed by Jing et al. (2025) for carbon dioxide heat pumps, have demonstrated significant potential to improve energy efficiency [22].

In this context, adjusting operating temperatures has been identified as a key strategy for improving energy efficiency. Adaptive temperature control allows reducing the use of heating and cooling systems while enhancing thermal comfort, as demonstrated by Forcada et al. (2021) in a study conducted in 5 nursing homes [23]. Bienvenido-Huertas et al. (2021) reached the same conclusion using a simulation of a residential building in Sevilla [24]. Sun et al. (2024) simulated the functioning of the Shandong (China) nursing home heating system and reached the conclusion that for each degree the temperature was reduced, energy consumption and carbon dioxide emissions decreased 2.3 kWh/m² and 0.8 kgCO₂/m², which represents a reduction of 11.1% and 11.9% respectively [25].

Other authors had similar results. Fong et al. (2023) estimated a 10% potential for energy-efficiency improvement through the use of self-cognizant prognostics for managing lighting, air conditioning, and food-preparation systems in a nursing home [26]. Zhou et al. (2024) estimated that the consumption of cooling energy can be reduced by 26.8% in a nursing home in Shanghai (China) through thermal landscape optimization strategies [27], proposing, among other measures, the use of an external shading system. Dursun & Aykut (2019) found that a nursing home with 200 residents in Istanbul (Türkiye) could operate exclusively on a hybrid renewable energy system [28], with an energy cost of 1306 $/kWh and CO₂ emissions of 5.44 kgCO₂/year.

Other authors have provided indicators of energy consumption in nursing homes. Lindberg et al. (2019) reported that the annual mean final energy consumption of 7 nursing homes in Norway was 260 kWh/m², with 140 kWh/m² for thermal energy and 120 kWh/m² for electrical power [29]. Kuzgunkaya (2019) concluded that the primary energy consumption of a nursing home in Türkiye was 271.91 kWh/m² a year [30]. Wang et al. (2017) found that a nursing home in Edinburgh (Britain) consumed 489 kWh/m², 377 kWh/m² for thermal energy, and 112 kWh/m² for electric power [31].

However, we observed that the mean electrical and thermal energy consumption, as well as the energy-related expenses and CO₂ emissions, in Spanish public nursing homes for the elderly had not yet been studied.

Research Objectives and Novelty

Currently, there are few publications regarding energy consumption indicators in nursing homes.

The objective of this study was to develop empirical predictive models of Energy Consumption, associated Costs, and CO₂ Emissions in nursing homes, based on readily available operational and structural variables, such as Useful floor area and the Number of Residents. Using data collected from 20 public nursing homes located in Extremadura (Spain), the study establishes statistically robust relationships that enable the evaluation of energy performance within this specific building type. This approach enables analysis of whether the energy consumption in a specific facility falls within an expected range or shows significant deviations [32].

The main novelty of this study lies in the development of benchmarking equations and reference indicators for nursing homes. Unlike more generic modeling approaches, the proposed models are derived from real operational data and rely on readily available variables, which enhances their practical applicability for facility managers and decision-makers. These results provide a quantitative basis for identifying deviations from expected energy consumption and supporting specific energy-efficiency and sustainability strategies and/or programs.

2. Materials and Methods

A quantitative analysis was conducted on 20 Spanish public nursing homes built or renovated between 1995 and 2009 in the region of Extremadura (Spain), spanning the period from 2019 to 2023. Concretely, the analysis was carried out on the nursing homes in

Table 1. List of nursing homes analyzed.

Code	Useful Floor Area (m2)	Number of Residents	Heating Degree Days	Cooling Degree Days
1	1500	50	1961	339
2	10,880	242	2214	253
3	3515	88	1835	399
4	1634	28	1897	304
5	8479	254	1897	304
6	1945	70	1961	339
7	1844	42	2386	256
8	4200	65	2126	229
9	9936	176	2311	266
10	1706	64	1945	370
11	2026	55	1975	234
12	9000	102	1937	465
13	1332	63	2498	206
14	8865	204	2104	273
15	3525	63	2076	408
16	1772	24	1945	370
17	3539	96	1945	370
18	2643	70	2104	273
19	5950	110	1961	339
20	1576	40	2066	278

, selected for having complete historical energy consumption records and representing the predominant operational and constructional profiles of public nursing homes in the region. Their useful floor area and number of residents ranged from 1332 to 10,880 m² and 24 to 254 residents, respectively, capturing the typical variability in building size, occupancy, and climatic conditions across Extremadura. Figure 1 shows a general scheme of the method and workflow followed.

Firstly, information about the nursing homes was collected, including useful floor area, number of residents, and year of construction, as well as any reforms that could influence the building’s energy consumption, such as changes to facilities or the thermal envelope. This information was obtained from the construction projects of each building. The directors of the nursing homes were asked to confirm that the number of residents remained unchanged, for example, due to organizational issues. This study adopts the term “Useful floor area” to denote the surface area available to residents after excluding the area occupied by internal partitions and building services.

The records of electricity and other fuel consumption were carefully analyzed to compare energy consumption across nursing homes. The energy consumption was calculated by converting the associated thermal energy into standardized units. Thermal demand was expressed as equivalent electrical energy using a normalized efficiency factor based on the performance coefficient (COP) of a standard air-cooled heat pump [20], defined as the ratio of useful thermal output to electrical energy input, enabling cross-system comparability. Based on this coefficient, the total energy consumption was calculated as

EC = EC_e + EC_t/β,

(1)

where EC, EC_e and EC_t stand for the total annual electrical and thermal energy consumptions (expressed in kWh), while β is a non-dimensional factor related with the provincial climate characteristics of the areas where the nursing homes are situated: 2.60, 2.65 and 2.70, for the Atlantic Northern, Mediterranean and Continental climatic regions [32].

The conversion of energy consumed to carbon dioxide emissions was performed using Equation (2). It is based on the direct use of final energy consumption by energy carriers, preserving the specific emission factors associated with each energy source and ensuring consistent estimation across different energy systems. The calculation is defined as follows:

E = EC_e × CF_e + EC_ng × CF_ng + EC_g × CF_g + EC_p × CF_p,

(2)

where E represents the annual carbon dioxide emissions (expressed in kgCO₂), EC_e, EC_ng, EC_g and EC_p represent the annual consumption for electricity, natural gas, diesel and propane, respectively (expressed in kWh), and CF_e, CF_ng, CF_g and CF_p are dimensionless conversion factors used, respectively, for electricity, natural gas, diesel and propane. The value of these conversion factors is 0.331, 0.252, 0.311 and 0.254 kgCO₂/kWh, respectively [33,34].

Subsequently, information on the regional weather was obtained, as weather plays a critical role in energy management and the accurate forecasting of energy behavior [35,36]. To do this, the network of meteorological stations from the Agroclimatic Information System for Irrigation was accessed [37]. This network has 47 weather stations in Extremadura. The average daily temperature between 2019 and 2023 was obtained for the different locations; in total, 30,681 data were available, from which the heating degree days, cooling degree days and total degree days using Equations (3), (4) and (5), respectively, were calculated:

$${\mathrm{HDD}}_{21} = {\displaystyle {\sum }_{\mathrm{i} = 1}^{\mathrm{n}}}({\theta }_{\mathrm{h}}- {\theta }_{\mathrm{i}}),$$

(3)

$${\mathrm{CDD}}_{23}={\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}({\theta }_{\mathrm{i}}-{\theta }_{\mathrm{c}}),$$

(4)

$$\mathrm{DD}={\mathrm{HDD}}_{21}+{\mathrm{CDD}}_{23},$$

(5)

where HDD₂₁ represents the heating degrees with base temperature 21 °C, CDD₂₃ represents the cooling degrees with base temperature 23 °C, θ_h and θ_c denote the base temperatures for heating (21 °C) and cooling (23 °C), respectively, θ_i is the average temperature on day i, and n is the number of days during the heating or cooling period [38]. These base temperatures define the thresholds at which buildings require heating or cooling and are consistent with the thermal setpoint ranges established by the Royal Decree 178/2021 of Spanish regulations [39].

The climate in Extremadura is Mediterranean, with some Atlantic influence, featuring warm summers and mild winters. The lowest average maximum daily temperatures (between 22 °C and 23 °C) correspond to a strip north of the province of Cáceres, which coincides with the most mountainous area of the Region. The rest of the territory of Extremadura has an average maximum temperature between 23 °C and 25 °C [40]. The difference between the average highest maximum temperature and the average lowest maximum temperature was 8.8 °C; between the highest average minimum temperature and the lowest average minimum temperature, the difference was 6.47 °C. This shows the variability of the existing maximum temperatures in Extremadura.

The application of linear regression has yielded favorable outcomes among statistical models, owing to its acceptable predictive performance and comparatively straightforward implementation relative to alternative approaches [41]. The correlations among the analyzed variables were examined using simple regression. This method conducts a linear regression analysis using the “least squares” method to fit a line through a set of observations. Previously, atypical values had been ruled out through Chauvenet’s Criterion technique.

Additionally, a multivariable regression analysis was conducted using all sample values. A significance level of 5% (corresponding to a 95% confidence interval) was adopted in this study, and Student’s t-test was applied. Residual analysis was conducted to evaluate the adequacy of the linear regression model.

Simple and multivariable regressions have expressions such as the type shown in Equations (6) and (7) respectively:

$$\mathrm{Y} = {\mathrm{a}}_0+ {\mathrm{a}}_1{\mathrm{X}}_1,$$

(6)

$$\mathrm{Y}={\mathrm{a}}_0+{\mathrm{a}}_1{\mathrm{X}}_1+{\mathrm{a}}_2{\mathrm{X}}_2+{\mathrm{a}}_{\mathrm{p}}{\mathrm{X}}_{\mathrm{p}},$$

(7)

where Y represents the response variable, X₁, X₂, … X_p are the predictor variables, with p indicating the overall quantity of predictors, and a₀, a₁, a₂ and a_p correspond to the regression coefficients.

The validity of the correlations was assessed through several statistical indicators, including the coefficient of determination (R²), adjusted coefficient of determination (adjusted R²), root mean square error (RMSE), mean absolute error (MAE) and relative error (RE). The expressions used for these calculations are provided below.

$${\mathrm{R}}^2=1-{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n}{\left({y_i-\widehat{y}}_i\right)}^2}{{\displaystyle {\sum }_{i=1}^n}{\left(y_i-\overline{\mathrm{y}}\right)}^2}} ,$$

(8)

$${\mathrm{R}}_{\mathrm{adj}}^2=1-(1-{\mathrm{R}}^2\left)\right(\mathrm{n}-1)/(\mathrm{n}-\mathrm{p}-1),$$

(9)

$$\mathrm{RMSE}=\sqrt{1/\mathrm{n}{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}}{\left({y_{\mathrm{i}}-\widehat{y}}_{\mathrm{i}}\right)}^2},$$

(10)

$$\mathrm{MAE}=1/\mathrm{n}{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}}\left|y_{\mathrm{i}}-\right.\left.{\widehat{y}}_{\mathrm{i}}\right|,$$

(11)

$$\mathrm{RE}=1/\mathrm{n}{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}}{\displaystyle \frac{\left|y_{\mathrm{i}}-\right.\left.{\widehat{y}}_{\mathrm{i}}\right|}{y_i}},$$

(12)

$$\mathrm{VIF}=1/\left(1-{\mathrm{R}}^2\right),$$

(13)

where y_i is the observed value of the dependent variable for the i-th observation, ${\widehat{\mathrm{y}}}_i$ is the value predicted by the i-th observation model, ${\overline{\mathrm{y}}}_{\mathrm{i}}$ is the mean of the observed values of the dependent variable, p is the number of independent variables and n is the number of observations in the sample.

The assumptions underlying the regression analysis were thoroughly evaluated, including linearity between the dependent and independent variables, independence of observations, normality and homoscedasticity of residuals, absence of influential points (outliers), and lack of multicollinearity, which occurs when strong linear correlations exist among the independent variables. Additionally, the presence of influential observations was examined using leverage and influence statistics, while multicollinearity among predictor variables was assessed using the Variance Inflation Factor (VIF), whose mathematical formulation is given in Equation (13) [42]. The findings were subsequently discussed and validated in collaboration with the engineers responsible for nursing homes.

3. Results

This section assesses the relationships between the variables and presents several linear regression models developed for this purpose. The outcomes are described in detail below.

3.1. Simple Linear Regression Models

The 9 relationships to be studied result from the combination of the 3 independent variables: useful floor area, number of residents, and degree days, and the 3 dependent variables: Energy Consumption, Energy Costs, and CO₂ Emissions. The selection criterion for the regression models was achieving a coefficient of determination (R²) greater than 0.80, as lower values were deemed to indicate a weak correlation between the variables. The simple regression plots corresponding to the paired analyses of variables for each predictive model are provided in Figure 2 and in Figures S1, S3, and S5 of the Supplementary Materials, respectively.

Useful floor area was found to correlate with Energy Consumption (R² = 0.8853), Energy Costs (R² = 0.8940) and CO₂ Emissions (R² = 0.8859). The more useful floor area that a nursing home has, the more energy it consumes and therefore, the greater the costs and emissions.

A correlation is observed relating resident count to Energy Consumption (R² = 0.9471), Energy Costs (R² = 0.9466), and CO₂ Emissions (R² = 0.9510). Energy consumption, associated Energy Costs, and CO₂ Emissions increase depending on the number of residents.

It was found that there is no correlation between the degree days and Energy Consumption (R² = 0.0848), Energy Costs (R² = 0.0987) and CO₂ Emissions (R² = 0.1022).

3.2. Multivariable Linear Regression Models

Given the strong correlation identified between the dependent and independent variables, several multivariable linear regression models were evaluated. The dependent variables considered were energy consumption (EC), energy costs (C), and CO₂ emissions (E), each analyzed in combination with the independent variables: useful floor area (S) and number of residents (N). The regression statistics for the proposed models are presented in

Table 2. Statistics of multivariable regression models for energy consumption.

	R2adj	RMSE	MAE	RE (%)	Max	Min	Average	Equation
Energy Consumption	0.9710	49 MWh	43 MWh	11%	1128 MWh	153 MWh	475 MWh	(14)
Energy Costs	0.9744	6766 €	5712 €	9%	168,331 €	26,450 €	74,229 €	(15)
CO2 Emissions	0.9742	24 kgCO2	18 kgCO2	8%	627 kgCO2	87 kgCO2	240 kgCO2	(16)

and

Table 3. Reference indicators.

Index	Units	Value
Annual energy consumption per m2	kWh/m2	111
Annual electrical energy consumption per m2	kWh/m2	63
Annual thermal energy consumption per m2	kWh/m2	47
Annual energy consumption per resident	kWh/resident	4983
Annual electrical energy consumption per resident	kWh/resident	2847
Annual thermal energy consumption per resident	kWh/resident	2136
Annual energy costs per m2	€/m2	17
Annual electrical energy costs per m2	€/m2	10
Annual thermal energy costs per m2	€/m2	7
Annual energy costs per resident	€/resident	779
Annual electrical energy costs per resident	€/resident	448
Annual thermal energy costs per resident	€/resident	331
Annual emission of carbon dioxide per m2	kgCO2/m2	56
Annual emission of carbon dioxide per m2 due to electrical energy consumption	kgCO2/m2	21
Annual emission of carbon dioxide per m2 due to thermal energy consumption	kgCO2/m2	35
Annual emission of carbon dioxide per resident	kgCO2/resident	2521
Annual emission of carbon dioxide per resident due to electrical energy consumption	kgCO2/resident	942
Annual emission of carbon dioxide per resident due to thermal energy consumption	kgCO2/resident	1578

.

Furthermore, the multicollinearity analysis showed that the VIF values for the independent variables under study were 4.81, which is below 5, indicating a moderate and acceptable level of multicollinearity [43]. Finally, the residual analysis revealed no significant deviations from normality or homoscedasticity, and no outliers with a disproportionate effect on the regression coefficients were identified. The underlying statistical process, along with the corresponding diagnostic analyses, is presented in the Supplementary Materials.

Equations (14)–(16) serve to determine the Energy Consumption, Energy Cost and CO₂ Emissions in a public nursing home, whose useful floor area (S) and number of residents (N) were in the range of 1332–10,880 m² and 24–254 residents. All the equations can be used in operational nursing homes, especially those built or renovated between 1995 and 2009. The equations are not applicable to new-design nursing homes because regulations aimed at enhancing the energy performance of buildings have evolved in recent years.

EC = 48 + 2.95N +0.03S,

(14)

C = 10,950 + 422.01N + 5.37S,

(15)

E = 13.87 + 1.58N + 0.02S,

(16)

Figure 3 shows annual Energy Consumption, Energy Costs, and CO₂ Emissions using both predicted and real models. The chosen vertical error bar corresponds to the overall relative error for each model.

Figure 4 shows the distributions of annual Energy Consumption, annual Energy Costs, and annual CO₂ Emissions by percentile. It reveals a clear upward trend, with more gradual growth in the lower and middle percentiles and a more pronounced increase in the upper percentiles. This pattern indicates greater variability among facilities with higher Energy Consumption, which is consistently reflected in the associated Energy Costs and CO₂ Emissions, and is linked to differences in operating conditions, building characteristics, or energy systems.

3.3. Reference Indicators

The percentage distribution of Energy Consumption was calculated to be 34% for electricity, 21% for natural gas, 36% for diesel, and 9% for propane. The percentage distribution of Energy Costs was calculated to be 57% for electricity, 13% for natural gas, 25% for diesel, and 5% for propane. The percentage distribution of CO₂ emissions was calculated as 37% for electricity, 18% for natural gas, 37% for diesel, and 8% for propane. Table 3 presents reference indicators related to energy consumption for the nursing homes that are the object of this study. The total Energy Consumption was broken down into electrical energy and thermal energy. The expenses related to energy and CO₂ Emissions have been included. Energy use was calculated by converting the corresponding thermal energy to its electrical equivalent. However, for calculating energy costs and carbon dioxide emissions, no such conversion was applied, as these values were derived directly from the thermal energy data. These indicators are useful for benchmarking against other nursing homes or buildings with other uses, such as educational or commercial buildings.

3.4. Cost Analysis

Figure 5 depicts the average, peak, and lowest costs of electricity and other fuels involved in the present study. Between 2019 and 2023, the price of electricity fluctuated between 90 €/MWh and 263 €/MWh, averaging 164 €/MWh. Natural gas fluctuated between 43 and 107 €/MWh, averaging 58 €/MWh. The maximum price of diesel was 104 €/MWh, the minimum was 45 €/MWh, and the average was 66 €/MWh. Propane fluctuated between 26 and 70 €/MWh, averaging 49 €/MWh during the study period.

4. Discussion

The equations proposed in this study are useful for several reasons. On the one hand, they allow the detection of inefficiencies in air-conditioning systems; on the other hand, they enable the comparison of different technologies or energy-efficiency strategies [44,45]. Having a reference to the energy cost for a nursing home improves financial planning and facilitates negotiating the most favorable energy purchase terms [46]. If a nursing home emits more CO₂ than the value obtained using the proposed formula in this study, this may indicate that its energy sources are highly polluting; therefore, it would make sense to consider renewing the facilities, for example, by installing renewable energy sources [47]. In addition, the proposed models can inform public policy by providing reliable estimates of projected energy consumption and costs, thereby enabling the identification of deviations and guiding the prioritization of energy-efficiency measures [48].

Reyna & Chester (2017) found that the demand for electricity and natural gas for residential buildings in Los Angeles could increase from 41% to 87% between 2020 and 2060 due to climate change [49]. This increase could be contained with appropriate policies aimed at improving the energy efficiency of buildings. In Spain, nursing homes are obliged to display an energy-efficiency label for the building when the total useful surface area exceeds 250 m² if publicly owned, and 500 m² if not, due to the exemplary role of public buildings [50,51]. Recently, the European Union has established new requirements for existing buildings to be renovated so as to turn them into almost null energy consumption buildings (nZEB) [52]. These standards establish that energy rehabilitation must be profitable from an economic point of view [53], and that the influence on the resilience of the building when facing extreme meteorological phenomena must be evaluated according to Sun et al. (2020) [54].

With improvements in building energy efficiency in recent years, the energy used to produce sanitary hot water has become more significant in the overall energy balance of buildings [55]. Heat losses from the pipes account for 38% of the total energy consumed by the hot water installation, according to results reported by Inamoto et al. (2018) in a nursing home in Japan [56]. However, Taxt Walnum et al. (2019) concluded that losses in the sanitary hot-water production systems of three nursing homes and three hotels in Norway were lower than expected [57]. The same authors concluded that more than 50% of the circulation systems do not work as planned [58]. The influence of building energy consumption management had already been shown by Garcia-Sanz-Calcedo et al. (2017) [59]. However, Martínez de Salazar et al. (2019) found that adequate energy management in a building was more effective than increasing users’ environmental awareness [60].

Roeger et al. (2022) found that, due to restrictions on gas imports from Russia following the Ukraine War, gas prices in the European Union had increased significantly since the summer of 2022 [61]. Zakeri et al. (2023) found that the European electricity market was highly exposed to gas price volatility [62]. The results obtained in this study highlight the practical applicability of the proposed models for energy management. In this context, the models allow estimating expected energy costs, supporting decision-making regarding energy supply strategies. The heat pump was identified as the most favorable option from an economic standpoint, followed by propane, natural gas, and diesel.

It was found that, similar to carbon dioxide emissions, buildings that used electricity as a thermal energy source, together with a heat pump, emitted the least CO₂ to the atmosphere, due to the high energy performance of this technology [63]. Natural gas was the next cleanest thermal energy source, followed by propane and diesel [64]. The construction of buildings, their operation, and the use of less clean energy sources have led to significant CO₂ emissions into the atmosphere in recent years [65]. Specifically, for every 10 buildings, approximately 900 kg of carbon dioxide is emitted into the atmosphere, which is around 39% of global CO₂ emissions every year [66].

Vergés et al. (2023) [20] reported coefficients of determination ranging from 0.46 to 0.80 in their linear regression models. Using the same dataset, Vergés et al. (2024) [19] developed an ANN-based model with strong predictive performance: a coefficient of determination of 0.95 and a relative error of 0.051. The two previous studies focused on forecasting the HVAC system’s energy demand throughout the cooling period. Therefore, they did not study HVAC energy consumption for the rest of the year and did not consider the energy consumption of other facilities: lighting, laundry, kitchen, medical devices, or sanitary hot water. The authors also did not propose reference indicators. Yiyu Ding et al. (2022) also concluded that artificial neural network models performed better in predicting heat consumption across 20 nursing homes [67]. However, compared to ANN-based models, the approach proposed in this study offers greater simplicity and direct applicability, providing interpretable reference values for comparative evaluation and practical decision-making [68]. This simplicity, both in its approach and in its computational requirements, facilitates integration into IoT devices, enabling implementation in operational environments without requiring advanced infrastructure [69]. Furthermore, by relying on readily available structural variables, such as Useful floor area and the Number of residents, the model exhibits reduced dependence on real-time data, thereby enhancing its robustness against potential failures in sensor systems [70].

Tartarini et al. (2017 and 2018) identified a lack of guidelines for thermal comfort and indoor environmental quality in the elderly care sector [71,72]. Subsequently, Forcada et al. (2020 and 2021) and Kainaga et al. (2022) developed studies that allowed us to know the preferred conditions both by the residents and the workers from a nursing home [73,74,75], which is especially important considering that the elderly spend 95% of their time indoors [76]. Forcada et al. (2020) found that, during winter, elderly residents tended to tolerate lower temperatures by adjusting their clothing, a behavior not observed among non-residents. Conversely, in summer, residents exhibited greater tolerance to high temperatures compared to non-residents [73].

The main limitation of this study is that its findings cannot be extrapolated to buildings constructed or extensively renovated after 2009, as energy efficiency regulations, such as the Energy Performance of Buildings Directive (EPBD), have since influenced building design practices and related parameters [77]. Furthermore, the sample size (20 nursing homes), while sufficient to identify consistent trends, may limit the generalizability of the results. On the other hand, although the MAE, RE, and RMSE were used to evaluate model performance, the validation approach is primarily internal. However, the diagnostic analysis performed, including residual assessment, outlier detection, and multicollinearity analysis, supports the internal consistency and stability of the resulting models. Finally, it should be noted that the models are based on aggregated data, which may not fully reflect differences in specific energy use across facilities.

Future studies should aim to provide a detailed analysis of nursing home energy consumption to inform the implementation of energy-saving measures.

5. Conclusions

The method used has proven effective for generating benchmark indicators and predictive equations from simple, readily available variables, facilitating their direct application in real-world operational contexts. In this regard, the developed models provide a practical benchmarking tool, enabling the identification of deviations from expected performance and supporting energy management and planning decisions in nursing homes.

The results show that the proposed models achieve high predictive power, with R_adj² values of 0.9710, 0.9744, and 0.9742 for Energy consumption, Energy Costs, and CO₂ Emissions, respectively. Average annual indicators of 111 kWh/m² and 4983 kWh/resident were identified for Energy Consumption, 17 €/m² and 779 €/resident for Energy Costs, and 56 kg CO₂/m² and 2521 kg CO₂/resident for CO₂ Emissions, thus confirming strong relationships with variables such as Useful floor area and the Number of Residents.

The applicability of the proposed approach depends on the characteristics of the buildings analyzed, particularly with regard to their construction period and operating conditions. Within these parameters, the models offer a simple and transparent solution to support energy efficiency and sustainability strategies in the elderly care sector.