Characterizing Healthcare Power Quality: Exploratory Findings from the MetaFridge Dataset

Written By: Allan Wasega (Data Analyst, nLine), Margaret Odero (Lead Data Analyst, nLine), and Rod Hinman (Principal, Aurora Research, LLC)

This post introduces the MetaFridge dataset - a high-resolution collection of voltage and frequency measurements from vaccine refrigerators deployed across public health facilities in Kenya - and illustrates its value through a case-study analysis of data from 2023. The dataset provides insight into the real electrical environments in which healthcare equipment operate, and our exploratory analysis shows that voltage and frequency were generally stable throughout the year. The findings demonstrate how the MetaFridge dataset can support context-aware equipment design and future research on energy conditions in healthcare settings.

Reliable electrical power is foundational to the effective functioning of healthcare systems. This is particularly the case in low- and middle-income countries (LMICs) where public health facilities frequently operate under resource and infrastructural constraints. In such environments, fluctuations in voltage and frequency, intermittent electricity supply, and a lack of backup generation options can have direct implications on medical equipment performance. Yet, the power systems that sustain this equipment are often poorly characterized, leading to limited evidence-based decision-making in energy and health infrastructure planning.

The MetaFridge dataset was developed precisely to address this gap. Originally collected by New Horizons, a division of Global Health Labs, the dataset was developed as part of research into the impact of power quality and availability on the operation of medical equipment in public health facilities. The MetaFridge dataset captures high-frequency measurements of grid voltage, current, frequency, and real power measured and reported by 270 vaccine-storage refrigerators installed in medical facilities across 45 out of the 47 counties in Kenya. These fridges captured granular (10-second resolution) data showing a real-world view of the electricity conditions under which frontline health care is delivered.

This blog post introduces this dataset, which nLine is making public on behalf of New Horizons. Through a case study analysis of data from 2023, we further characterize the temporal and spatial patterns of two electrical parameters: voltage and frequency. By linking these parameters to geographic metadata, we illustrate how power conditions vary across counties and over time, and highlight the types of insights such granular data can enable.

About the MetaFridge Dataset

Scope and Coverage

The dataset spans multiple years from Q2 2017 through Q2 2025, capturing measurements at each installed unit on key electrical parameters including line current, line frequency, line voltage, and real power (Figure 1). Each measurement record is associated with a timestamp of measurement (measuredDtm) and record creation (createdDtm), facilitating temporal continuity and metadata tracking. As the MetaFridges act as distributed sensor nodes in health-facility settings, the dataset represents one of the few publicly available repositories where end-use electrical behavior is tied to healthcare-infrastructure contexts.

Figure 1. Sample MetaFridge dataset header: The dataset contains seven features - fridgeID, measuredDtm, createdDtm, line_current, line_frequency, line_voltage, and line_real_power. — **Figure 1. Sample MetaFridge dataset header:** The dataset contains seven features - fridgeID, measuredDtm, createdDtm, line_current, line_frequency, line_voltage, and line_real_power.

Data Enrichment and Geospatial Linking

The public dataset is hosted on nLine’s MetaFridge public data page and made available as downloadable Parquet files, thereby supporting open access and reproducible research. In addition to the raw electrical measurements, the dataset can be enriched by linking FridgeIDs to their geographic locations. The fridges’ location information is available in a supplementary file - MF Power Kenya Location - which contains data including the county, sub county, and electoral ward in which a unit was installed. This spatial anchoring enables cross-region comparisons of power-system behavior in health-care contexts.

The rest of this blog post proceeds as follows. First, we present a case study analysis demonstrating how data from the MetaFridge dataset can be used to identify and characterize power quality patterns across healthcare facilities. Next, we discuss the broader significance of these patterns for healthcare operations, emphasizing how variations in voltage and frequency can influence equipment reliability and ultimately the effectiveness of cold-chain systems. Finally, we conclude by inviting collaboration around open data use, reproducible research, and applied analytics aimed at improving power quality and reliability in health facilities.

Case Study Analysis: Exploratory Insights from the 2023 MetaFridge Subset

To illustrate how the MetaFridge dataset can be used to characterize power quality patterns across healthcare facilities, we analyzed one full year of data (January–December 2023). The 2023 dataset contained approximately 770 million observations across all active refrigerators, each linked to its respective county and sub-county. This high-density dataset enables both temporal and spatial characterization of voltage and frequency behavior at the county and sub-county levels.

Descriptive statistics and spatial patterns

To understand typical operating conditions, we first examined the distributions of voltage and frequency across all measurements. Overall, the average voltage and frequency was 237.42 V and 50.14 Hz, respectively. These values are both close to Kenya’s nominal grid standards (240 V and 50 Hz) (Figures 2 and 3). When examining the distributions at the county level, the mean, median, and interquartile ranges (25th–75th percentiles) for both voltage and frequency fall well within their respective nominal tolerance bands. The boxplots also show the full spread of observed values, with occasional extremes that occur infrequently relative to the tightly clustered bulk of measurements. In general, the findings indicate stable voltage and frequency conditions across the counties.

Figure 2. Average voltage by county, sorted by mean voltage: For all counties, the median and interquartile range fall within the nominal ±10% voltage band (216–264 V). This indicates that sampled facilities in the counties operate under voltage conditions aligned with expected standards, although some variation in the voltage spreads occur across regions. — **Figure 2. Average voltage by county, sorted by mean voltage:** For all counties, the median and interquartile range fall within the nominal ±10% voltage band (216–264 V). This indicates that sampled facilities in the counties operate under voltage conditions aligned with expected standards, although some variation in the voltage spreads occur across regions.

Figure 3. Average frequency by county, sorted by mean frequency: Across counties, nearly all frequency measurements lie within the typical operating band of 49.5–50.5 Hz. The whiskers reflect the full extent of observed values, capturing infrequent dips or peaks, but the bulk of readings remain close to the nominal 50 Hz target. — **Figure 3. Average frequency by county, sorted by mean frequency:** Across counties, nearly all frequency measurements lie within the typical operating band of 49.5–50.5 Hz. The whiskers reflect the full extent of observed values, capturing infrequent dips or peaks, but the bulk of readings remain close to the nominal 50 Hz target.

Temporal patterns in voltage quality

Beyond understanding the overall distribution of voltage across counties, examining temporal patterns reveals how voltage behaves throughout the day and across the calendar year. Time-of-day and seasonal variations can reflect underlying dynamics in electricity demand, grid loading, and distribution network performance, all of which can directly influence equipment operating conditions. Figures 4 and 5 illustrate these temporal dimensions: Monthly aggregates (Figure 4) reveal year-round stability, while hourly patterns (Figure 5) capture the predictable daily fluctuations driven by demand cycles.

Figure 4. Average monthly voltage: Average monthly voltages are highly stable within the nominal voltage range (216 V to 264 V). — **Figure 4. Average monthly voltage:** *Average monthly voltages are highly stable within the nominal voltage range (216 V to 264 V).*

Figure 5. Daily voltage variation (Hourly aggregates by day of week): Voltage levels are highest at midnight (0000 hours) and lowest at 1600 hours, illustrating the classic load-driven stress pattern common in distribution networks. — **Figure 5. Daily voltage variation (Hourly aggregates by day of week):** *Voltage levels are highest at midnight (0000 hours) and lowest at 1600 hours, illustrating the classic load-driven stress pattern common in distribution networks.*

Voltage reliability assessment

To evaluate voltage suitability for sensitive equipment, we applied data-driven thresholds derived from established electrical standards. With Kenya’s nominal voltage of 240 V and an allowable deviation of ±10%, we classified voltage readings as follows:

Usable: 216–264 V (within tolerance)

Unusable (undervoltage): Below 216 V but above 24 V

Unusable (overvoltage): Above 264 V

Outage: Below 24 V (less than 10% of nominal)

Figure 6. Proportion of usable vs. unusable voltage measurements: Across most counties, over 90% of voltage readings fall within the usable range of 216–264 V. Garissa shows the lowest proportion of usable measurements at 77.8%, while the remaining counties exhibit only small shares of overvoltage or undervoltage observations. — **Figure 6. Proportion of usable vs. unusable voltage measurements:** Across most counties, over 90% of voltage readings fall within the usable range of 216–264 V. Garissa shows the lowest proportion of usable measurements at 77.8%, while the remaining counties exhibit only small shares of overvoltage or undervoltage observations.

The 2023 data reveals three key patterns. First, voltage levels across counties align well with Kenya's nominal standards, with mean voltages falling within the accepted ±10% band. Second, frequency measurements show similarly consistent results, with nearly all counties exhibiting distributions within the typical 49.5–50.5 Hz operating band. Third, threshold-based assessment confirms this stability: most counties show over 90% usable voltage readings, with Garissa being the notable exception at 77.8%. Taken together, these findings demonstrate how granular data can provide a detailed view of the electrical conditions under which healthcare facilities operate, with findings from Kenya as an example.

Averages can obscure critical vulnerabilities: The case of MetaFridge `2891381510072434752` in Elgeyo Marakwet County

New Horizons installed seven MetaFridges across three regions in Elgeyo Marakwet County: Keiyo South, Keiyo North, and Marakwet East. Cumulatively, these devices recorded a county-wide average voltage of 243.73 V in 2023, which is within Kenya’s nominal voltage range (see Figure 2 above). However, Elgeyo Marakwet’s box plot also shows that the county had the highest recorded voltage value, which leads to the question: was this an isolated incident or an indicator of voltage issues obscured by averages? To investigate this question further, we isolated the MetaFridge with this highest recorded voltage value in the county (Figure 7 below).

Figure 7: Individual MetaFridge voltage statistics. MetaFridge 2891381510072434752 in Keiyo South recorded the highest voltage value of 285.5 V. Its average voltage was still within the nominal range, although higher than the aggregate of 243.7 V. — **Figure 7: Individual MetaFridge voltage statistics.** MetaFridge `2891381510072434752` in Keiyo South recorded the highest voltage value of 285.5 V. Its average voltage was still within the nominal range, although higher than the aggregate of 243.7 V.

A time series plot of the MetaFridge’s voltage readings reveals some insightful patterns (Figure 8). Importantly, a vast majority of readings (99.47%) are within the nominal voltage range, meaning the facility enjoys good power quality on average. However, we also see a number of sustained overvoltage readings (0.5% of data points), which points to some power quality issues at the hospital. Such overvoltages can be especially concerning in a hospital running electric equipment with strict input voltage requirements. For example, one of the available electrocardiographs in Kenya is the BeneHeart 12, which has an input voltage range of 100 to 240 VAC with a tolerance of +/-10%. This device is, therefore, likely to malfunction when connected directly to the same power supply as MetaFridge 2891381510072434752 without any voltage stabilizers. Figure 8 further reveals some multi-day power outages (such as between May 31st and June 10th, and between November 21st and December 3rd). Thus, we see evidence of concerning power quality and reliability issues that might have otherwise been missed by focusing solely on the average county-wide data.

Figure 8: Time series scatter plot of recorded voltages at the facility hosting MetaFridge 2891381510072434752. While the vast majority (99.47%) of voltage readings are within the nominal range, we see a number of undervoltage (0.03%) and overvoltage (0.50%) readings as well as multi-day outages (red arrows). Due to the volume of data, the chart above shows every third data point, equivalent to a 30 second sampling rate. — **Figure 8: Time series scatter plot of recorded voltages at the facility hosting MetaFridge** `2891381510072434752`. While the vast majority (99.47%) of voltage readings are within the nominal range, we see a number of undervoltage (0.03%) and overvoltage (0.50%) readings as well as multi-day outages (red arrows). Due to the volume of data, the chart above shows every third data point, equivalent to a 30 second sampling rate.

Sector-Wide Relevance across the Health-Energy Ecosystem

The patterns identified in the MetaFridge dataset highlight the value of high-resolution electrical data for understanding the operating environments of healthcare equipment in Kenya.

Policymakers and public health planners can use the dataset to ground energy-related decisions in real-world conditions. For example, our analysis reveals that while most counties maintain voltage and frequency within nominal ranges, the captured variations in stability and occasional excursions can help identify where targeted infrastructure investments or backup-power strategies may be most beneficial.

For biomedical engineers and equipment designers, the dataset provides a detailed picture of actual operating electrical environments in Kenya. The generally stable voltage and frequency distributions, punctuated by occasional spikes or dips, give engineers a realistic sense of what equipment must tolerate in practice. These insights can inform decisions on voltage tolerance ranges, compressor design, and component durability.

For researchers and innovators, the dataset creates opportunities for deeper exploration of facility-level power characteristics in low-resource settings. The observed stability in voltage and frequency measurements makes the dataset well suited for evaluating equipment performance under typical operating conditions and designing predictive or simulation models. The data can also be extended to support cross-disciplinary work spanning electrical engineering, data analytics, and global health.

Conclusion

This exploration of the MetaFridge dataset demonstrates the value of high-resolution, facility-level electrical measurements for understanding the operating environments of healthcare equipment in Kenya. By examining spatial and temporal patterns in voltage and frequency across public health facilities, we show how granular data can illuminate the day-to-day conditions under which cold-chain devices function. These insights provide important context for energy-related activities within the healthcare space, such as evaluating equipment performance and designing technologies that are well matched to real-world electrical conditions.

The public release of the MetaFridge dataset marks an important step toward evidence-based decision-making at the intersection of energy and health. We encourage researchers, analysts, policymakers, and technologists to explore this resource - whether to validate reliability metrics, build predictive models, assess equipment risk, or investigate new dimensions of healthcare energy security.

Access the full MetaFridge dataset at nline.io/public-data/metafridge. The New Horizons team also made available a variety of documents related to vaccine cold-chain, including system designs, technical guidance, and field reports here). For questions or collaboration opportunities, please contact info@nline.io and rod@auroraresearch.com.