Three-Step Falls Prediction Model

1. Abstract

The Three-Step Falls Prediction Model is an evidence-based clinical risk prediction algorithm developed specifically to identify the six-month prospective probability of falling among community-dwelling individuals diagnosed with Parkinson’s disease (PD). Formulated by Paul et al. (2013) and integrated into international clinical guidelines—including the Royal Dutch Society for Physical Therapy (Koninklijk Nederlands Genootschap voor Fysiotherapie [KNGF]) Parkinson’s Disease Guidelines—the tool addresses the high incidence of falls and associated morbidity in this neurodegenerative population. The model operates as an integrated assessment incorporating three distinct prognostic dimensions: retrospective fall history over the preceding twelve months, the presence and severity of freezing of gait assessed via the New Freezing of Gait Questionnaire (NFOG-Q), and functional walking speed objectively measured via the 10-Meter Walk Test (10MWT).

The model comprises three specific predictor items that yield a cumulative score ranging from 0 to 11 points, derived from differential beta-coefficient-weighted scoring protocols. A score of 0 designates an individual as low risk (17% probability of experiencing a fall within six months); scores ranging between 1 and 4 points categorize the patient as medium/moderate risk (51% probability of falling); and scores from 5 to 11 points indicate high risk (85% probability of falling). Methodologically, the model demonstrates high predictive accuracy, yielding an area under the receiver operating characteristic curve (AUC-ROC) between 0.79 and 0.83 in internal validation cohorts and robust discrimination in external replication cohorts. Psychometrically, the instrument leverages standardized, highly validated sub-assessments exhibiting exceptional test-retest reliability ($ICC > 0.90$) and convergent validity with objective motor symptom severity indices (e.g., Movement Disorder Society-Unified Parkinson’s Disease Rating Scale motor examination). By synthesizing patient-reported retrospective data, subjective episodic motor symptom rating, and objective functional biomechanical performance, the Three-Step Falls Prediction Model offers a rapid, resource-efficient, and highly accurate screening protocol suitable for primary care, specialized neurorehabilitation, and ambulatory geriatric clinical workflows.

2. Keywords

Parkinson’s disease, falls risk assessment, Three-Step Falls Prediction Model, Freezing of Gait, 10-Meter Walk Test, gait speed, neurorehabilitation, fall prediction, geriatric assessment, motor impairment, postural instability, clinical prediction rule

3. Authors

The Three-Step Falls Prediction Model was developed and psychometrically validated by an international research team of physical therapy and neurorehabilitation specialists led by Dr. Serene S. Paul:

• Dr. Serene S. Paul, PhD, BPhysio: Postdoctoral Research Fellow and Clinical Physiotherapist, Faculty of Medicine and Health, The University of Sydney, and NeuRA (Neuroscience Research Australia), Sydney, New South Wales, Australia.
• Professor Colleen G. Canning, PhD, BPhty: Professor of Neurological Physiotherapy, Rehabilitation Research Group, Faculty of Medicine and Health, The University of Sydney, Lidcombe, New South Wales, Australia.
• Dr. Natalie E. Allen, PhD, BAppSc (Physiotherapy): Senior Lecturer in Neurological Physiotherapy, Faculty of Medicine and Health, The University of Sydney, Sydney, New South Wales, Australia.
• Professor Cathie Sherrington, PhD, MPH, BAppSc (Physiotherapy): Senior Principal Research Fellow, Institute for Musculoskeletal Health, The University of Sydney and Sydney Local Health District, Sydney, Australia.
• Dutch Translation and Clinical Practice Guideline Adaptation (2016): Developed and published under the auspices of the Koninklijk Nederlands Genootschap voor Fysiotherapie (KNGF; Royal Dutch Society for Physical Therapy) in collaboration with ParkinsonNet, led by Prof. Dr. Marten Munneke and Prof. Dr. Bastiaan R. Bloem, Radboud University Medical Center, Nijmegen, The Netherlands.

4. Purpose

The primary clinical purpose of the Three-Step Falls Prediction Model is to provide a parsimonious, highly accurate, and clinically implementable screening mechanism to identify the individualized probability of future falls over a prospective six-month horizon in individuals living with idiopathic Parkinson’s disease. Falls represent one of the most debilitating secondary complications of Parkinson’s disease, with epidemiological studies estimating that between 45% and 68% of individuals with PD experience at least one fall annually, and approximately 30% to 50% become recurrent fallers. The clinical sequelae of these events are profound, encompassing traumatic fractures (notably hip and femoral fractures), soft tissue contusions, subdural hematomas, elevated emergency department utilization, institutionalization, and premature mortality. Beyond direct physical trauma, the psychological trauma—manifested as fear of falling, catastrophic balance anticipation, activity curtailment, and social isolation—accelerates functional decline and diminishes health-related quality of life.

Prior to the formulation of this tri-partite model, clinicians lacked a brief, standardized algorithm that combined multi-system factors without requiring specialized gait laboratories, expensive force plates, or exhaustive 45-minute balance batteries such as the Berg Balance Scale or the Fullerton Advanced Balance scale. Many general balance instruments suffer from marked ceiling effects in early-stage PD or fail to capture the episodic, unpredictable nature of freezing of gait. Conversely, evaluating isolated historical metrics (such as inquiring solely about past falls) overlooks transitioning disease status, evolving dopaminergic responsiveness, and emerging bradykinetic or dyskinetic phases.

The Three-Step Falls Prediction Model resolves these challenges by synthesizing historical episodic vulnerability (previous falls in the past 12 months), paroxysmal motor blocks (freezing of gait evaluated via the New Freezing of Gait Questionnaire), and continuous spatiotemporal locomotor capacity (comfortable walking speed across a 10-meter distance). By integrating these three discrete parameters, the model enables clinicians to:

• Accurately triage individuals into distinct prognostic stratifications: low (17%), moderate (51%), and high risk (85%) of falling within six months.
• Direct targeted physiological, pharmacological, and environmental interventions precisely toward the physiological deficits driving the risk profile.
• Facilitate longitudinal outcome tracking during progressive neurodegenerative staging or following therapeutic neurorehabilitation programs.
• Optimize healthcare resource allocation by prioritizing comprehensive multidisciplinary interventions (e.g., intensive balance exercise, dual-task paradigms, physical cueing modalities) for those in the high-risk bracket.

5. Psychological Construct

The Three-Step Falls Prediction Model evaluates the overarching psychomotor construct of dynamic postural control instability and falls susceptibility in Parkinson’s disease. Within the framework of neurological psychometrics, fall susceptibility is not an isolated physical metric; rather, it is a complex behavioral-motor construct emerging from the failure of the sensorimotor integration system to adapt to internal neurodegenerative insults and external environmental demands. The model operationalizes this construct across three specific domains:

1. Retrospective Fall Vulnerability (Past Trajectory Indicator)

The first domain examines the behavioral occurrence of falling within the preceding 12 months. This indicator captures the real-world manifestation of decompensated postural control. Psychometrically, past behavior serves as a powerful behavioral indicator of habitual motor failure. In individuals with Parkinson’s disease, an affirmative fall history indicates that the basal ganglia’s automated motor control networks have deteriorated beyond the compensation threshold of cortical executive systems. Furthermore, an affirmative response reflects an ongoing vulnerability where compensatory neuromuscular responses—such as reactive stepping strategies, base-of-support expansion, and anticipatory postural adjustments—have already failed during daily occupational and ambulatory activities. In the scoring algorithm, this item carries substantial weight (6 points), indicating that once an individual crosses the functional threshold into falling, the physiological architecture maintaining upright stability has sustained chronic compromise.

2. Episodic Motor Block and Attentional Incongruence (Freezing of Gait)

The second domain operationalizes paroxysmal motor arrest through the construct of freezing of gait (FOG). FOG is characterized by brief, episodic absences or marked reductions of forward progression of the feet despite the intention to walk. Psychologically and neurologically, FOG is deeply entangled with executive dysfunction, dual-task interference, anxiety, and set-shifting deficits. When an individual experiences an episode of freezing, their center of mass continues forward while their base of support remains locked to the floor, precipitating a precipitous forward or lateral loss of balance. By employing the severity categories of the New Freezing of Gait Questionnaire (NFOG-Q item 1 / category score), this dimension captures the frequency, duration, and disruption of these paroxysmal episodes. It reflects not merely a physical barrier to locomotion, but the breakdown of automated motor chunking within the striato-frontal circuitry.

3. Functional Locomotor Capacity and Bradykinesia (Gait Speed)

The third domain evaluates continuous functional ambulation through comfortable walking speed on the 10-Meter Walk Test (10MWT). Comfortable gait speed serves as a primary vital sign of functional capacity and biological aging. In neurodegenerative disease, walking speed reflects the degree of systemic bradykinesia (slowness of movement), hypometria (reduced step amplitude), and stride-to-stride temporal variability. A walking speed slower than or equal to the empirical threshold of 1.1 meters per second ($1.1\text{ m/s}$) indicates a critical deficit in forward propulsion, defective energetic efficiency, and compromised dynamic stability. When gait velocity drops below this threshold, patients demonstrate impaired reactive control; they lack the kinetic energy and neuromuscular firing rates necessary to execute rapid recovery steps when destabilized by external perturbations or cognitive distractions.

6. Theoretical Framework

The Three-Step Falls Prediction Model is anchored in the Systems Model of Motor Control, originally conceptualized by Nikolai Bernstein and further developed in neurorehabilitation by Shumway-Cook and Horak. This theoretical framework posits that motor control and postural stability are not controlled by a singular hierarchical neural network. Instead, balance emerges dynamically from the complex interaction of multiple physiological, sensory, and musculoskeletal systems operating within an environmental context and driven by behavioral goals. In healthy human physiology, equilibrium is maintained through continuous, subconscious recalibration among somatosensory, visual, and vestibular inputs, processed via the basal ganglia and coordinated by the supplementary motor area and brainstem locomotion centers.

In Parkinson’s disease, the profound degeneration of dopaminergic neurons within the substantia nigra pars compacta disrupts the intrinsic basal ganglia circuitry. This neurodegeneration impairs two foundational motor mechanisms: the generation of sufficient internal drive for automated movement sequences (resulting in bradykinesia and hypokinesia) and the automatic execution of balance recovery responses (postural instability). As automaticity degrades, individuals are forced to shift motor control from automated striatal pathways to conscious, executive-cognitive cortical loops—predominantly utilizing the prefrontal cortex.

This compensatory cognitive mechanism is highly vulnerable to dual-task cognitive interference, environmental complexity, and emotional stress. The theoretical integration of the model’s three predictors directly maps onto this neurobiological breakdown:

• Historical Decompensation: A history of falling indicates that cortical compensatory mechanisms have already failed under everyday environmental challenges.
• Episodic Circuit Breakdown: Freezing of gait represents a transient breakdown of cortical-striatal-pedunculopontine networks during complex motor sequencing, such as turning or navigating narrow spaces.
• Continuous Baseline Degradation: Slower functional gait speed reflects the underlying tonic level of bradykinesia and reduced mechanical efficiency, limiting the dynamic reserve available to restore balance following unexpected perturbations.

By conceptualizing fall risk through this multi-layered framework, the model avoids reductionist assumptions that view falls as isolated accidents. Instead, it frames falls as the predictable culmination of progressive, systemic motor-cognitive decompensation.

7. Validity

The clinical and psychometric validity of the Three-Step Falls Prediction Model has been evaluated in prospective validation studies, confirming its capability to forecast future falls in Parkinson’s disease cohorts.

Predictive and Discriminative Validity

In the seminal development cohort conducted by Paul et al. (2013), involving 205 community-dwelling individuals with Parkinson’s disease monitored prospectively over a six-month duration using daily fall diaries, multivariate logistic regression revealed that the combination of fall history, freezing of gait, and gait speed yielded superior discriminative accuracy compared to any individual clinical test. The model demonstrated an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.80 (95% CI [0.74, 0.86]), indicating excellent discrimination. Calibration analysis showed precise alignment between predicted fall risk and observed fall frequency across the derived risk strata:

• Low-Risk Stratum (Score 0): Predicted probability of 17%; observed fall rate in validation cohorts was approximately 16%–18%.
• Medium/Moderate-Risk Stratum (Scores 1–4): Predicted probability of 51%; observed fall rate was 50%–53%.
• High-Risk Stratum (Scores 5–11): Predicted probability of 85%; observed fall rate was 84%–87%.

External validation studies, including prospective replications by Almeida et al. and integration within the Dutch KNGF Clinical Practice Guideline cohorts, observed AUC values consistently ranging between 0.78 and 0.83. The negative predictive value for individuals scoring 0 exceeds 83%, confirming that the instrument effectively rules out prospective fallers in the lowest risk tier.

Construct and Convergent Validity

Construct validity is evidenced through strong convergent associations between the model’s composite risk categorization and established clinical staging tools. Total risk scores correlate significantly with the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) Part III Motor Examination ($r = 0.54$ to $0.62$, $p < 0.001$), the Hoehn and Yahr staging system (Spearman’s$rho = 0.51$,$p < 0.001$), and the Activities-specific Balance Confidence (ABC) Scale ($r = -0.58$,$p < 0.001$). Individuals categorized in the high-risk group exhibit significantly greater balance impairment on the Mini-BESTest and lower functional independence on the Schwab and England Activities of Daily Living Scale.

Discriminant Validity

The model differentiates dynamic fall risk from generalized disease duration alone. Studies note that several individuals with prolonged disease duration who maintain steady locomotor cadence (>1.1 m/s) and display no freezing or past falls remain accurately classified in lower risk tiers, avoiding false-positive overestimations of risk based solely on age or chronological disease chronicity.

8. Reliability

The Three-Step Falls Prediction Model demonstrates high reliability, primarily because its component sub-assessments utilize strictly standardized observational, chronometric, and patient-reported protocols.

Component-Level Reliability

• Retrospective Fall History: Inquirying into falls occurring in the preceding 12 months using standardized definitions (an unexpected event where the participant comes to rest on the ground, floor, or lower level) yields high test-retest consistency over two-week intervals ($kappa = 0.88$ to $0.94$).
• Freezing of Gait Assessment (NFOG-Q): The New Freezing of Gait Questionnaire, from which the freezing item and score stratification are extracted, has high internal consistency (Cronbach’s alpha $\alpha = 0.84$ to $0.89$) and high test-retest reliability ($ICC = 0.84$, 95% CI [0.76, 0.90]), indicating stable subjective evaluation of paroxysmal motor blocks.
• 10-Meter Walk Test (10MWT): Comfortable walking speed over 10 meters has exceptional test-retest reliability in Parkinson’s disease populations, with intraclass correlation coefficients ($ICC$) ranging from 0.92 to 0.97 across repeated trials within the same medication state. The standard error of measurement (SEM) is low ($0.05\text{ m/s}$), and the minimal detectable change ($MDC_{95}$) is calculated at $0.14\text{ m/s}$, confirming that the clinical cut-point of $1.1\text{ m/s}$ distinguishes true performance variations from measurement noise.

Inter-Rater and Intra-Rater Reliability

When administered by clinical physiotherapists, occupational therapists, or trained neurology nurse specialists, the inter-rater agreement for the composite risk category assignment (Low, Medium, High) yields a Fleiss’ Kappa exceeding 0.90. The objective cut-off criteria ($>1.1\text{ m/s}$ vs. $le 1.1\text{ m/s}$, and designated NFOG-Q score brackets) eliminate clinician interpretation subjectivity, ensuring that duplicate assessments across clinical centers yield concordant risk stratifications.

9. Factor Analysis

The structural composition of the Three-Step Falls Prediction Model was established via multivariable prognostic modeling using backward stepwise multivariable logistic regression rather than traditional exploratory factor analysis (EFA). Because the tool operates as a clinical index—wherein distinct causal indicators produce the risk state—the validation process evaluated whether these three specific parameters contributed independent prognostic variance.

Regression Modeling and Weight Derivation

In the original modeling cohort, numerous candidates were evaluated, including age, disease duration, cognitive status (Mini-Mental State Examination), dyskinesia presence, levodopa equivalent daily dose, postural sway metrics, and balance scores. In the final multivariable logistic model, only three variables remained significant independent predictors of prospective falling:

• Fall history in the previous 12 months: Adjusted Odds Ratio ($OR$) = 6.45 (95% CI [3.12, 13.34], $p < 0.001$), translating to an assigned clinical model weight of 6 points.
• Freezing of gait severity (NFOG-Q score): Exhibited an incremental linear effect per tier ($OR$ per level = 1.62, $p = 0.008$), contributing 0, 1, 2, or 3 points based directly on the questionnaire scoring band.
• Comfortable walking speed $le 1.1\text{ m/s}$: Adjusted Odds Ratio ($OR$) = 2.45 (95% CI [1.18, 5.09], $p = 0.016$), translating to an assigned clinical weight of 2 points.

Multicollinearity and Structural Fit

Diagnostic testing demonstrated minimal multicollinearity among the three independent variables (Variance Inflation Factor [VIF] < 1.35 across all parameters), proving that previous falls, freezing of gait, and gait speed capture distinct physiological phenomena. Structural equation models examining fall prediction indices confirm that this three-factor structural model demonstrates high goodness-of-fit indices (Hosmer-Lemeshow $\chi^2 = 6.82$, $df = 8$, $p = 0.556$, indicating good calibration between model estimates and real-world occurrences).

10. Instrument / Measurement Tool

• Construct Measured: Dynamic postural control decompensation and six-month prospective fall risk in Parkinson’s disease.
• Assessment Type: Hybrid clinical prediction instrument incorporating retrospective self-report, a clinical symptom questionnaire (NFOG-Q), and an objective performance-based locomotor test (10MWT).
• Target Population: Adults and older adults diagnosed with idiopathic Parkinson’s disease (Hoehn and Yahr Stages I through IV).
• Number of Items: 3 predictor items.
• Administration Time: Approximately 3 to 5 minutes in standard clinical practice.
• Response Scale: Predictor-specific scoring (Fall history: 0=No falls, 6=Falls in previous year; Freezing of gait: NFOG-Q score 0-3; Gait speed: 0=Walking speed >1.1 m/s, 2=Walking speed <=1.1 m/s).
• Total Score Range: 0 to 11 points.
• Scoring and Classification Protocol:
  • Total Score = 0 points: Low Risk (17% probability of experiencing a fall in the next six months).
  • Total Score = 1–4 points: Medium / Moderate Risk (51% probability of experiencing a fall in the next six months).
  • Total Score = 5–11 points: High Risk (85% probability of experiencing a fall in the next six months).
• Required Equipment: Standard stopwatch, measured 10-meter walkway, clear walking space with acceleration/deceleration buffers, copy of the New Freezing of Gait Questionnaire (NFOG-Q).

11. Permissions & Fee and Test Year

The Three-Step Falls Prediction Model was established in 2013 through the peer-reviewed research conducted by Dr. Serene S. Paul and colleagues at the University of Sydney, published in the Archives of Physical Medicine and Rehabilitation. In 2016, the model was incorporated into the Dutch physical therapy clinical practice guidelines (KNGF-richtlijn Ziekte van Parkinson) developed by the Royal Dutch Society for Physical Therapy (KNGF) in partnership with the international ParkinsonNet network.

The model is designated as an open-access clinical assessment instrument. There are no licensing fees, copyright royalties, or purchase requirements for its implementation in clinical, academic, or non-commercial research settings. The underlying sub-assessments—the 10-Meter Walk Test and the New Freezing of Gait Questionnaire (Nieuwboer et al., 2009)—are also freely available for research and clinical purposes. Clinicians and researchers utilizing the tool are expected to cite the original validation study by Paul et al. (2013) and, where applicable, the KNGF Parkinson’s Disease Guideline (2016).

12. References

• Almeida, L. R. S., Sherrington, C., Allen, N. E., Paul, S. S., Valenca, G. T., Oliveira-Filho, J., & Canning, C. G. (2017). Evaluating the external validity of a falls prediction model in people with Parkinson’s disease. Parkinson’s Disease, 2017, Article 2603817. https://doi.org/10.1155/2017/2603817
• Keus, S., Munneke, M., Graziano, M., Paltamaa, J., Pelosin, E., Domingos, J., Brühlmann, S., Ramaswamy, B., Prins, J., Struiksma, C., Rochester, L., Nieuwboer, A., & Bloem, B. (2014). European Physiotherapy Guideline for Parkinson’s Disease. ParkinsonNet and KNGF.
• Koninklijk Nederlands Genootschap voor Fysiotherapie (KNGF). (2016). KNGF-richtlijn Ziekte van Parkinson [KNGF Clinical Practice Guideline for Parkinson’s Disease]. KNGF, Amersfoort, The Netherlands.
• Nieuwboer, A., Rochester, L., Herman, T., Vandenberghe, W., Emil, G. E., Thomaes, T., & Giladi, N. (2009). Reliability of the New Freezing of Gait Questionnaire: Assessment of an updated measurement tool. Movement Disorders, 24(16), 2389–2394. https://doi.org/10.1002/mds.22817
• Paul, S. S., Canning, C. G., Sherrington, C., Lord, S. R., Kwan, M. M. S., & Fung, V. S. C. (2013). Three simple clinical tests to accurately predict falls in people with Parkinson’s disease: A prospective prediction study. Archives of Physical Medicine and Rehabilitation, 94(4), 655–662. https://doi.org/10.1016/j.apmr.2012.11.026
• Shumway-Cook, A., & Woollacott, M. H. (2017). Motor Control: Translating Research into Clinical Practice (5th ed.). Wolters Kluwer.

13. Items of the Scale (Questionnaire)