A prospective, randomised controlled trial will be conducted with 230 community-dwelling participants with Parkinson's disease (Figure 1).

To be included in the study participants must meet the following inclusion criteria: (i) a diagnosis of idiopathic Parkinson's disease; (ii) aged 40 years or over; (iii) able to walk independently with or without a walking frame; (iv) fallen at least once in the past 12 months or found to be at risk of falls on assessment; (v) have adapted to their current anti-Parkinsonian medication for at least 2 weeks.

Participants will be excluded if they have a Mini-Mental State Examination score of < 24, or suffer from unstable cardiovascular disease, or suffer from other uncontrolled chronic conditions that would interfere with the safety and conduct of the training and testing protocol or interpretation of the results.

All volunteers will be screened by a physiotherapist. Medical clearance will be required from each participant's medical practitioner to certify him/her as able to participate in moderate-intensity semi-supervised exercise before being accepted into the trial.

The study protocol has been approved by The University of Sydney Human Research Ethics Committee (HREC Number 11-2007/10487), and Sydney West Area Health Service Human Research Ethics Committee (HREC2008/9/4.13 (2865) AU RED 08/WMEAD/250) and written informed consent will be obtained from all participants.

All participants will undergo two measurements: one on entry to the study (pre-test) and one after the 6-month intervention period (post-test). All pre-test measures will be assessed prior to randomisation. Cognitive function will be assessed with the Mini Mental State Examination 21. Measurements will be conducted one hour after each participant's last dose of L-dopa, thus each participant will be tested in their "on" phase.

After completion of the pre-test assessment, participants will be formally entered into the study and randomized to intervention or control groups. Randomization will be stratified by falls history (0–10 falls in the previous 12 months/more than 10 falls in the previous 12 months) using a computer-generated random number schedule with variable block sizes of 2–6. Randomization will be performed centrally by an investigator not involved in recruitment or assessments.

The control group will receive their usual care from their medical practitioner and community services. In addition, standardised falls prevention advice in the form of a booklet will be provided.

The primary outcome measure will be falls:

Falls will be assessed by comparing the number of falls in intervention and control groups. The proportion of fallers in each group will also be compared. Falls will be recorded by the use of a "falls diary". All participants will receive monthly calendars on entry to the study, with instructions to record the following events: number of falls, visits by or to nursing and allied health personnel, general practitioner or specialists appointments and hospitalisations. Participants will be asked to return the completed calendar monthly in pre-paid envelopes to the research personnel who are unaware of group allocation. All participants will also be telephoned monthly to record any changes in medications, use of health resources and verify any falls details (including how and where the fall occurred, injuries suffered, medical intervention required and limitations to activity as a result of a fall).

To evaluate the cost effectiveness of the exercise program the following measures will be collected:

Costs of the intervention: the costs of implementing the exercise program will be obtained from trial records and research group financial records using actual costs when available. Costs associated with developing and evaluating the exercise program will be excluded from the analysis.

Costs of health services used for managing fall-related injuries: use of public and private healthcare resources will be recorded as part of the monthly telephone calls, over the 6-month intervention period.

The secondary outcome measures listed below will be collected on entry to the study and at the end of the 6-month intervention period by an assessor unaware of group allocation. The order of measurements will be standardized. Participants will be instructed not to inform the assessors of their intervention status, and all home exercise equipment will be removed or concealed prior to the post-assessment.

Secondary outcome measures are:

(i) Parkinson's Disease Falls Risk Score: Falls risk status will be determined using the Parkinson's disease-specific algorithm developed in our recent large prospective cohort study 3. This involves weighted contributions from neurological and physiological function measures including gait freezing, impaired balance and lower limb weakness.

(ii) Maximal muscle strength of the knee extensor muscles of each leg will be tested using a strain gauge 25.

(iii) Balance in standing, using the coordinated stability test which requires participants to accurately adjust their position in a steady and coordinated manner when their centre of mass is near the limits of their base of support 26.

(iv) Short Physical Performance Battery, including tests of walking, balanced standing and sit to stand 27.

(v) Freezing of gait, using the Freezing of Gait Questionnaire 28.

(vi) Falls Efficacy Scale International, a falls efficacy questionnaire 29.

(vii) Health and Well-being, using the SF12v2™.

(viii) Habitual Physical Activity Questionnaire to quantify the type and amount of regular physical activity.

(ix) PDQ-39, a Parkinson's disease specific quality of life questionnaire 30.

(x) Positive and Negative Affect, using the Positive and Negative Affect Schedule (PANAS) 31.

Adverse events (defined as a significant injury or medical event that causes the participant to seek attention from a health professional or limit their activities) will be monitored and recorded throughout the study during the monthly telephone calls to each participant.

The number of falls per person-year will be analysed using negative binomial regression to estimate the difference in fall rates between the two groups, adjusted for previous multiple faller status and other confounding variables if required 32. The proportion of fallers between groups will be compared using the relative risk statistic. Between-group comparisons of final test performance for the continuously-scored outcome measures will be made using General Linear Models (ANCOVA) controlled for pre-test performance. Baseline continuously-scored data will be compared between groups using the student t-test. Ordinally-scored data will be analysed for between group differences using the non-parametric Mann Whitney U statistic. An intention-to-treat approach will be used for all analyses.

Economic analysis will be undertaken in the manner used in previous falls prevention studies 3334. The total costs in intervention and control groups (including the cost of the exercise program, and costs of health and community service contacts) and falls rates, will be used to calculate an incremental cost per fall prevented in the intervention group compared with the control group. Bootstrapping will be used to estimate a distribution around costs and health outcomes and to calculate confidence intervals around the cost-effectiveness ratio; one-way sensitivity analyses will be conducted around key parameters.

As fall rates will be compared between groups using incident rate ratios (IRR) from negative bionomial regression models 32 we conducted an analysis of statistical power using the nbpower command in the STATA software package. A total of 230 participants (115 per group) will be required to provide 80% power to detect as significant, at the 5% level, a 30% lower rate of falls for exercise subjects than control subjects (i. e. IRR = 0.70). We assumed the control group rate of falls would be 1 fall/person month over the 6-month follow-up period, a conservative estimate as the prospective cohort study 3 reported a falls rate of 1.6 falls/person month in a 12-month follow up of 113 people with PD.

The sample size of 230 will also be sufficient to detect between-group differences of 23% of control group values in the coordinated stability (balance) measure (assuming a control group mean of 18 and SD of 14) and 11% for knee extension strength (assuming a control group value of 16 kg and SD of 6). Both calculations are based on values found among fallers with PD 3 assuming 80% power of a two-sided test at the 5% level, a correlation between pre-and post measures of 0.7, and 15% dropouts.