Factors associated with the melanoma diagnostic interval in Ontario, Canada: a population-based study

Abstract

Background

Protracted times to diagnosis of cancer can lead to increased patient anxiety, and in some cases, disease progression and worse outcomes. This study assessed the time to diagnosis for melanoma, and its variability, according to patient-, disease-, and system-level factors.

Methods

This is a descriptive, cross-sectional study in Ontario, Canada from 2007–2019. We used administrative health data to measure the diagnostic interval (DI)–and its two subintervals–the primary care subinterval (PCI) and specialist care subinterval (SCI). Multivariable quantile regression was used.

Results

There were 33,371 melanoma patients. The median DI was 36 days (interquartile range [IQR]: 8–85 days), median PCI 22 days (IQR: 6–54 days), and median SCI 6 days (IQR: 1–42 days). Increasing comorbidity was associated with increasing DI. Residents in the most deprived neighbourhoods and those in rural areas experienced shorter DIs and PCIs, but no differences in SCI. There was substantial variation in the DI and SCI across health regions, but limited differences in the PCI. Finally, patients with a history of non-melanoma skin cancer, and those previously established with a dermatologist experienced significantly longer DI, PCI, and SCI.

Discussion

This study found variability in the melanoma DI, notably by system-level factors.

Introduction

The global incidence and mortality of melanoma are increasing. In 2020, approximately 325,000 people were diagnosed with melanoma worldwide [1]. In Canada, melanoma is the seventh most common cancer diagnosed among men and women, and the fourth most common cancer diagnosed in those aged 15 to 49 years [2]. Melanoma is primarily caused by exposure to ultraviolet radiation, and the rising incidence of melanoma has been attributed to increasing exposure to the sun or artificial ultraviolet radiation sources [3]. The rising incidence of melanoma is also thought to be due to an increase in awareness and detection of thin, treatable lesions [4–6]. Nevertheless, until recently, the mortality rate of melanoma has not declined, suggesting that the incidence of thicker, prognostically unfavourable lesions was not decreasing [6, 7]. Detection of melanoma at an early stage is associated with increased survival. Recent data from Canada reports five-year relative survival ranging from 99% for localised disease to 14% for distant disease [8].

Timely detection of melanoma is dependent on the awareness of symptoms by patients, their families, and their physicians. To improve the timeliness of a melanoma diagnosis, we must understand factors associated with a protracted diagnostic period. The diagnostic interval (DI) is the time between the first presentation to the healthcare system with signs and symptoms suggestive of cancer and diagnosis of the cancer, and can be divided into several sub-intervals, notably the primary care subinterval and the specialist care subinterval [9]. Studies on other cancers have identified variations in the length of the DI according to various patient, disease, and system-level factors [10–14]. Few studies have been conducted quantifying the melanoma DI [15–20], and only one study, to our knowledge, has been conducted in Canada [21] examining the length of time to diagnose melanoma.

We need to understand factors associated with a protracted DI to improve the timeliness of a melanoma diagnosis. The purpose of this study was to quantify the DI, and its subintervals in patients diagnosed with melanoma, and to identify independent patient-, disease-, and system-level factors associated with prolonged intervals.

Methods

Study population

This was a population-based descriptive cross-sectional study of melanoma patients diagnosed between January 1, 2007 and October 31, 2019 in the province of Ontario, which is the most populous province in Canada, accounting for almost 40% of its inhabitants. Ontario has a publicly funded universal healthcare system, which allows all residents to access medically necessary hospital and physician services.

Patients with invasive melanoma were identified in the Ontario Cancer Registry by the ICD-O-3 definition: C44 with histologies 8720–8790. Patients were excluded if they had less than 60 months of continuous Ontario Health Insurance Plan (OHIP) coverage preceding diagnosis and 6 months post diagnosis, were missing age or sex, were less than 20 years of age as of their index date, were a non-Ontario resident at diagnosis, or had a concurrent cancer. From this eligible study cohort, we excluded those who we were unable to assign an index contact date (defined later).

Ethics approval

This study was approved by the Health Sciences Research Ethics Board at Queen’s University. Participant consent has been waived as the research falls under Article 5.5 A of the Tri-Council Policy Statement 2 (TCPS 2). The study was reported in accordance with the STROBE (STRengthening the Reporting of OBservational studies in Epidemiology) and RECORD (REporting of studies Conducted using Observational Routinely-collected health Data) statements [22, 23].

Data sources and variables

ICES is an independent, non-profit research institute whose legal status under Ontario’s health information privacy law allows it to collect and analyse health care and demographic data, without consent, for health system evaluation and improvement. Databases accessed at ICES include the Ontario Cancer Registry (OCR), the Canadian Institute of Health Information Discharge Abstract Database (CIHI-DAD), Same Day Surgery (SDS), National Ambulatory Care Reporting System (NACRS), the ICES Physician Database, the OHIP claims database, census data, and Registered Persons Database (RPDB). Datasets were linked using unique encoded identifiers and analysed at ICES.

The OCR is a passive provincial cancer registry known for its high degree of population coverage [24], with a large-scale validation study showing capture rates over 96% for invasive melanomas diagnosed between 2007 and 2009 [25]. Date of diagnosis, histology, anatomic location, and limited stage data were available from OCR. Stage data from OCR was supplemented using pathology reports abstracted for a previously developed retrospective population-based cohort study (CIHR MOP 137022) of a 65% random sample of all melanomas diagnosed in Ontario between January 1, 2007 to December 31, 2012 [26]. CIHI-DAD and SDS provided hospitalisation information, including admission and discharge dates, as well as diagnoses and procedures. NACRS provided data from outpatient clinic visits and emergency department visits such as admission dates, diagnoses, and procedures conducted. The ICES physician database provided physician specialty as well as physician practice postal code. OHIP provided the diagnosis and procedure codes, as well as the physician rendering the service and date of service. Census data provided population counts. RPDB provided demographic data, including age, sex, local health integration network (LHIN), area-level rurality as calculated by the Rurality Index of Ontario (RIO) [27, 28], and area-level socioeconomic status, as calculated by the material deprivation component from the Ontario Marginalisation Index [29], all of which were assigned as of the index contact date. LHINs were anonymized.

Comorbidity was assigned using the Johns Hopkins® ACG System v10.0.1 (build 879) Aggregated Diagnosis Groups (ADGs), using data from the 0–24-month period prior to the index contact date [30]. The ACG system uses records in OHIP, CIHI DAD, SDS, and NACRS. Both minor and major comorbidities were examined. Concurrent cancer was defined as a cancer, other than the index melanoma, that was diagnosed between the index date and 6-months post-melanoma diagnosis and found in OCR. History of non-melanoma skin cancer (NMSC) in the 5-years prior to the index contact date was ascertained using an algorithm developed by Chan [31], which was shown to have 82.3% sensitivity and 92.9% specificity, as they are not contained within OCR. ‘Established with dermatology’ was defined as 3 or more visits with a dermatologist in the 5 years prior to the index contact date. Dermatologist density was calculated at the county level per 100,000 population. Dermatologists seen by the patients in the cohort made up the numerator, and the denominator was total county size. Dermatologist density was calculated separately for each year, to account for fluctuations in both population size and dermatologist supply. Each patient was assigned a dermatologist density value according to year of diagnosis.

Determination of the diagnostic interval

The diagnostic interval (DI) is the time (in days) between the first melanoma-related presentation to the healthcare system and melanoma diagnosis. Detailed methods to identify the first melanoma-related encounter, also known as the index contact visit, using administrative health data are described elsewhere [32]. Briefly, we created a list of ‘probable’ melanoma-related diagnosis and procedure codes by examining those that were more common in the three months prior to diagnosis compared to the 12-to-15-month period prior to diagnosis, which we considered as the background, or control, period. Diagnosis and procedure codes were kept if they had face validity, as determined by our team of melanoma-related experts (TH, YA, NLH, FW, HL). We then grouped like-codes together into 22 distinct categories (Appendix A). Lookback periods were determined for each category separately using statistical control charts [33–35]. Due to the long duration of our study period, and to account for healthcare variations over time, we defined lookback periods over three time periods – those diagnosed in 2007–2011, 2012–2015, and 2016–2019. We used four established rules to assign four potential lookback period cut-points [35] and assessed each category’s signal strength at these cut-points (Appendix B). The signal strength is a measure of the proportion of visits in the lookback period that exceeds that which is expected based on the weekly rate of visits in the control period. We removed encounter categories that did not achieve a signal strength of 80%, apart from non-melanoma skin conditions, which were considered to have high face validity but did not reach that threshold. For each patient, we then collected all relevant encounters within their respective lookback periods and identified each patient’s earliest encounter. This earliest encounter was the index contact date unless it was the result of a referral. In those cases, we added the most recent visit with the referring physician and that became the index date.

The diagnosis date was defined based on biopsy and procedure code (Appendix C).

The DI was calculated as the number of days between the diagnosis date and the index date.

Melanoma-related specialist

A melanoma-related specialist was defined as: a dermatologist, a plastic surgeon, a general surgeon, a neurologist/neurosurgeon, or an otolaryngologist. This choice was based on the specialties that the initial physician was most likely to refer to in our data, and based on consultation with our experts. A primary care provider (PCP) in Ontario can acquire a sub-specialty designation in dermatology, allowing them to accept referrals from other PCPs in Ontario. In this situation, the PCP with a sub-specialty designation in dermatology can bill OHIP for a consultation. These consults associated with a melanoma-related diagnostic code were also included.

Primary care and specialist care subintervals

In defining the subintervals, we used principles articulated in the Aarhus statement [9]. The PCI was defined as the period between the first visit with a non-melanoma specialist and either the first visit with a melanoma specialist or diagnosis if no melanoma specialist was seen. If there was no evidence of a melanoma-related encounter with a non-melanoma specialist, the patient was not assigned a PCI.

The SCI was defined as the period between the first visit with a melanoma-related specialist and the diagnosis. If a patient did not see a melanoma-related specialist, they were not assigned an SCI.

Statistical analyses

We described the demographic characteristics, disease characteristics, and the lengths of the DI, PCI, and SCI for the study population. The interval lengths were positively skewed, as such, quantile regression was used to examine both the 50^th percentile (median) and 90^th percentile, where parameter estimates are the difference in days between variable categories at the chosen percentile [36]. Studying the 50th percentile provides information about differences in the majority experience while examining the 90th percentile provides information on extreme waits. To explore whether these factors were independently associated with the DI, PCI, and SCI, we performed mutually adjusted multivariate analyses. We considered differences of more than 1 week as clinically meaningful. We did not adjust for stage because it was missing for 45% of patients. We conducted a sensitivity analysis on the staged subset to assess whether the additional adjustment for stage explained our results. All statistical analyses were conducted using SAS Enterprise Guide version 7.1 (SAS Institute Inc., Cary, NC).

Results

Study population

The final study cohort consisted of 33,371 patients (Fig. 1), of which, 81% of patients had a PCI, 82% had an SCI, and 63% had both a PCI and SCI. Patient and disease characteristics of the whole cohort, along with patients in the PCI and SCI cohorts, can be found in Table 1. Overall, there were more males than females (54.6%) and the mean age was 63.6 years (standard deviation 15.7 years). Thirty-eight percent of patients had no major ADGs, and 30.6% had one major ADG. More patients lived in the least deprived neighbourhoods compared to the most deprived (26.1% vs. 13.2%), and most patients lived in urban areas (88.5%). The most common melanoma histology was superficial spreading melanoma at 34.9%, and 40.6% had melanoma NOS. The most common body sites of melanoma were the extremities and trunk (44.4% and 32.1%, respectively). Stage was missing for 45.1% of our cohort, but among those with stage information, 58.7% had stage I. Seventeen percent of our cohort had a history of NMSC, and 20% were established with dermatology prior to their index date. A majority of the cohort lived in counties with a dermatologist density greater than 1 per 100,000 population (67.1%), however, almost 20% of patients lived in counties with no dermatologists.

Fig. 1. Cohort creation flowchart.

Flowchart describing the process of cohort creation.

Table 1.

Patient-, disease-, and system-level characteristics for the overall cohort, and the primary care and specialist care subinterval cohorts.

	Overall cohort N = 33,371	Primary care subinterval (n = 26,987)	Specialist care subinterval (n = 27,264)
Characteristic	%	%	%
Patient factors
Sex
Male	54.6	53.7	54.7
Female	45.4	46.3	45.3
Age Mean (SD)	63.6 (15.7)	63.1 (15.8)	63.9 (15.7)
20–45	13.5	14.1	13.1
46–55	16.1	16.7	15.9
56–65	22.2	22.5	21.9
66–75	22.9	22.4	22.9
76–85	18.2	17.3	18.7
>85	7.0	6.8	7.5
Comorbidity
Minor ADG
0–1	10.2	10.6	10.0
2	10.0	10.4	9.7
3	11.7	12.4	11.4
4	13.1	13.4	12.8
5	12.7	12.8	12.8
6	11.4	11.3	11.4
7	9.7	9.3	9.8
8	7.7	7.2	7.9
9+	13.7	12.6	14.4
Major ADG
0	38.0	39.9	37.1
1	30.6	30.6	30.4
2	17.5	16.5	17.7
3+	14.0	12.9	14.7
Material deprivation quintile
1 – least deprived	26.2	25.3	26.3
2	22.9	22.8	23.0
3	20.2	20.6	20.0
4	16.8	17.0	16.7
5 – most deprived	13.2	13.5	13.3
Missing	0.7	0.7	0.7
Rurality
Rural	11.6	11.9	10.3
Non-major urban	28.1	28.9	26.3
Major urban	60.4	59.2	63.4
History of NMSC
Yes	17.3	13.2	18.6
No	82.7	86.8	81.4
Established with dermatology
Yes	19.7	12.2	22.7
No	80.3	87.8	77.3
Disease factors
Histology
Superficial spreading melanoma	34.9	35.8	35.2
Nodular melanoma	11.5	12.0	11.9
Lentigo maligna melanoma	8.4	7.6	9.4
Acral lentiginous melanoma	0.9	0.8	1.0
Other	3.6	3.6	3.8
NOS	40.6	40.3	38.9
Location
Extremities	44.4	45.1	44.3
Trunk	32.1	32.5	31.6
Face	12.1	11.5	13.2
Scalp and neck	6.7	6.3	7.1
NOS	4.7	4.7	3.7
Stage
I	32.2	32.3	32.0
II	13.0	13.4	13.3
III	7.1	7.4	7.1
IV	2.6	2.6	2.2
Missing	45.1	44.3	45.4
System factors
County-level dermatologist density per 100,000 population
0	18.9	19.7	16.8
>0, ≤1	13.9	14.2	13.7
>1, ≤2	32.2	32.9	32.5
>2	34.9	33.2	37.1
LHIN
A	7.1	6.1	8.0
B	9.6	9.4	9.5
C	8.8	8.5	9.9
D	5.3	5.3	5.0
E	10.2	10.9	8.5
F	6.5	6.4	6.9
G	5.5	5.7	5.4
H	11.7	12.0	12.5
I	6.8	6.9	6.9
J	1.7	1.7	1.5
K	3.3	3.5	3.5
L	13.3	13.3	13.6
M	5.7	5.9	4.7
N	4.6	4.6	4.3

SD Standard Deviation, ADG Adjusted Diagnostic Groups, NMSC Non-Melanoma Skin Cancer, NOS Not Otherwise Specified, LHIN Local Health Integration Network.

Patient characteristics differed slightly among those in the whole cohort, PCI cohort, and the SCI cohort. Of note, patients with a PCI were less likely to have a history of NMSC or be established with a dermatologist prior to their melanoma diagnosis compared to the whole cohort (13.2% and 12.2% vs. 17.3% and 19.7%, respectively). Additionally, a slightly higher proportion of patients with an SCI lived in higher dermatologist-density counties (37.1% in SCI cohort living in counties with dermatologist density >2/100,000 vs. 33.2% in PCI cohort and 34.9% in DI cohort).

Factors associated with the duration of the DI, PCI, and SCI

Table 2 presents the median and 90^th percentile of the DI, PCI, and SCI by patient and disease characteristics. The overall median DI was 36 days (Interquartile range [IQR]: 8–85 days) and 90% of patients were diagnosed within 142 days. The median PCI was 22 days (IQR: 6–54), and median SCI was 6 days (IQR: 1–42). Table 3 summarizes the results of the adjusted quantile regression models evaluating the median and 90^th percentile of the DI, PCI, and SCI. See Supplementary Appendix D for a summary of main results for DI.

Table 2.

Diagnostic interval, primary care subinterval, and specialist care subinterval lengths according to patient-, disease-, and system-level factors.

	Diagnostic interval (DI) (n = 33,371)		Primary care subinterval (PCI) (n = 26,987)		Specialist care subinterval (SCI) (n = 27,264)
Characteristic	DI 50th percentile, days	DI 90th percentile, days	PCI 50th percentile, days	PCI 90th percentile, days	SCI 50th percentile, days	SCI 90th percentile, days
Overall	36.0	142.0	22.0	100.0	6.0	97.0
Patient factors
Sex	p = 1.00	p = 0.65	p = 0.10	p = 0.007	p = 0.004	p < 0.001
Male	36.0	141.0	22.0	98.0	7.0	100.0
Female	36.0	142.0	23.0	103.0	5.0	92.0
Age	p < 0.001	p = 0.02	p = 0.05	p < 0.001	p < 0.001	p < 0.001
20–45	34.0	148.0	23.0	113.0	2.0	79.0
46–55	30.0	133.0	22.0	104.0	2.0	74.0
56–65	34.0	138.0	21.0	98.0	3.0	88.0
66–75	38.0	143.0	22.0	99.0	8.0	104.0
76–85	42.0	143.0	22.0	94.0	11.0	113.0
>85	45.0	145.0	24.0	92.0	14.0	114.0
Comorbidity
Minor ADG	p < 0.001	p < 0.001	p = 0.16	p = 0.29	p < 0.001	p < 0.001
0–1	28.0	124.0	20.5	96.0	1.0	65.0
2	31.0	130.0	21.0	99.0	3.0	73.0
3	33.0	139.0	22.0	105.0	3.0	79.0
4	35.0	142.0	22.0	99.0	5.0	92.0
5	39.0	142.0	24.0	100.0	6.0	95.0
6	36.0	144.0	22.0	100.0	5.0	101.0
7	41.0	146.0	23.0	98.0	9.0	108.0
8	39.0	146.0	22.0	99.0	8.0	111.0
9+	49.0	152.0	22.0	104.0	16.0	121.0
Major ADG	p < 0.001	p < 0.001	p = 0.17	p = 0.02	p < 0.001	p < 0.001
0	30.0	128.0	21.0	98.0	2.0	71.0
1	36.0	141.0	22.0	100.0	5.0	98.0
2	42.0	150.0	22.0	101.0	10.0	113.0
3+	49.0	162.0	23.0	106.0	15.0	130.0
Material deprivation quintile	p = 0.08	p = 0.003	p < 0.001	p < 0.001	p = 0.15	p = 0.27
1 – least deprived	38.0	146.0	24.0	106.0	7.0	100.0
2	37.0	144.0	23.0	103.0	5.0	98.0
3	36.0	141.0	22.0	98.0	5.0	94.0
4	35.0	139.0	21.0	99.0	6.0	94.0
5 – most deprived	35.0	132.0	20.0	91.0	6.0	98.0
Rurality	p < 0.001	p = 0.91	p < 0.001	p = 0.19	p = 0.001	p = 0.41
Rural	27.0	140.0	17.0	97.0	3.0	94.0
Non-major urban	34.0	142.0	21.0	99.0	7.0	99.0
Major urban	39.0	142.0	24.0	102.0	6.0	98.0
History of NMSC	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001
Yes	58.0	177.0	28.0	112.0	25.0	144.0
No	33.0	133.0	21.0	98.0	3.0	80.0
Established with dermatology	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001
Yes	52.0	164.0	30.0	120.0	21.0	127.0
No	34.0	135.0	21.0	97.0	3.0	80.5
Disease factors
Histology	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001
Superficial spreading melanoma	35.0	136.5	22.0	99.0	5.0	86.0
Nodular melanoma	34.0	122.0	20.0	90.0	6.0	85.0
Lentigo maligna melanoma	50.0	154.0	27.0	103.0	14.0	113.0
Acral lentiginous melanoma	53.0	146.0	27.5	100.5	27.5	112.0
Other	42.0	134.5	23.0	98.0	9.0	101.0
NOS	35.0	148.0	22.0	105.0	5.0	104.0
Location	p < 0.001	p < 0.001	p < 0.001	p = 0.17	p < 0.001	p < 0.001
Extremities	35.0	137.0	22.0	102.0	3.0	90.0
Trunk	33.0	138.0	21.0	97.0	5.0	93.0
Face	50.0	153.0	27.0	104.0	13.0	110.0
Scalp and neck	45.0	147.0	25.0	100.0	12.0	105.0
NOS	31.0	160.0	18.0	102.0	16.0	133.0
Stage	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001
I	34.0	134.0	22.0	101.0	4.0	87.0
II	33.0	123.0	20.0	90.0	4.0	89.0
III	30.0	122.0	19.0	90.0	3.0	79.0
IV	21.0	120.0	14.0	84.0	7.0	101.0
Missing	41.5	157.0	25.0	105.0	8.0	108.0
System factors
County-level dermatologist density per 100,000 population	p < 0.001	p = 0.14	p < 0.001	p = 0.23	p < 0.001	p = 0.07
0	30.0	140.0	20.0	97.0	4.0	97.0
>0, ≤1	38.0	148.0	22.0	98.0	9.0	100.0
<1, ≤2	38.0	141.0	22.0	101.0	8.0	94.0
>2	37.0	141.0	24.0	102.0	2.0	99.0
LHIN	p < 0.001	p = 0.009	p < 0.001	p < 0.001	p < 0.001	p < 0.001
A	38.0	137.0	23.0	101.0	4.0	105.0
B	33.0	143.0	24.0	108.0	1.0	91.0
C	42.0	141.0	24.0	96.0	9.0	98.5
D	34.5	142.0	21.0	92.0	8.0	100.0
E	29.0	142.0	19.0	101.0	2.0	102.0
F	35.0	137.0	25.0	100.0	2.0	87.5
G	43.0	141.0	19.0	87.0	20.0	105.0
H	42.0	150.0	24.0	103.0	7.0	98.0
I	45.0	156.0	24.0	102.0	14.0	106.0
J	28.0	128.0	15.0	100.0	14.0	76.0
K	37.0	130.0	22.0	91.0	7.0	93.0
L	37.0	135.0	23.0	102.0	5.0	91.0
M	27.0	138.0	20.0	99.0	1.0	103.0
N	29.0	148.0	19.0	112.0	4.0	85.0

PCI Primary Care subInterval, SCI Specialist Care subInterval, ADG Adjusted Diagnostic Groups, NMSC Non-Melanoma Skin Cancer, NOS Not Otherwise Specified, LHIN Local Health Integration Network.

p-values are calculated from unadjusted quantile regression.

Bold values highlight a significant p-value less than 0.05

Table 3.

Results from adjusted^a quantile regression for the diagnostic interval, primary care subinterval, and specialist care subinterval.

	Diagnostic interval				Primary care subinterval				Specialist care subinterval
	50th percentile		90th percentile		50th percentile		90th percentile		50th percentile		90th percentile
Characteristic	B	Confidence interval (95%)	B	Confidence interval (95%)	B	Confidence interval (95%)	B	Confidence interval (95%)	B	Confidence interval (95%)	B	Confidence interval (95%)
Intercept	31.2	27.6–34.7	126.8	116.9–136.7	23.5	21.2–25.8	99.3	90.3–108.3	1.0	−0.5–1.5	68.9	61.4–76.5
Patient factors
Sex
Male	−1.1	−2.6–0.4	−4.3	−8.5–−0.1	−1.0	−1.9–−0.1	−2.1	−5.8–1.6	0.0	−0.2–0.2	−5.9	−9.5–−2.2
Female	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Age
20–45	2.7	0.3–5.0	16.1	9.2–23.1	2.2	0.6–3.9	15.9	10.0–21.9	0.0	−0.2–0.2	−0.8	−7.3–5.8
46–55	0.4	−1.6–2.4	0.3	−5.9–6.5	0.4	−1.2–1.9	4.3	−1.3–9.9	0.0	−0.2–0.2	−6.7	−12.8–−0.7
56–65	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
66–75	1.6	−0.7–3.9	−1.3	−7.4–4.8	−0.1	−1.5–1.3	−3.8	−9.0–1.5	0.3	−0.2–0.9	2.1	−3.4–7.7
76–85	0.3	−2.2–2.8	−12.2	−18.2–−6.1	−1.5	−3.1–0.1	−10.3	−15.8–−4.8	0.3	−0.4–1.1	−2.9	−8.3–2.5
>85	1.8	−1.5–5.1	−15.2	−22.6–−7.8	−0.4	−2.6–1.7	−13.2	−19.6–−6.8	0.0	−1.6–1.6	−4.5	−13.5–4.4
Comorbidity Minor ADG
0–1	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
2	3.1	0.2–6.0	10.4	1.1–19.7	1.3	−0.7–3.4	5.5	−2.8–13.8	0.0	−0.3–0.3	2.2	−3.8–8.3
3	2.9	0.0–5.7	12.2	2.6–21.8	0.7	−1.2–2.6	5.9	−2.3–14.1	0.0	−0.3–0.3	1.2	−6.0–8.3
4	5.7	2.7–8.8	11.6	3.4–19.7	1.2	−1.0–3.4	4.0	−4.4–12.3	0.0	−0.5–0.5	7.4	0.2–14.7
5	6.8	3.8–9.7	8.1	−0.8–17.1	1.9	−0.2–3.9	0.9	−5.8–7.7	0.0	−0.4–0.4	4.8	−2.4–12.0
6	3.2	0.0–6.5	9.7	0.5–18.9	0.5	−1.7–2.6	2.4	−5.3–10.1	0.0	−0.4–0.4	11.8	3.4–20.3
7	5.3	2.0–8.6	6.0	−3.2–15.2	0.4	−2.0–2.7	−0.4	−8.7–7.8	2.0	0.9–3.1	14.0	5.5–22.4
8	3.9	0.6–7.3	6.0	−2.4–14.3	−1.1	−3.4–1.3	−1.6	−9.4–6.2	0.0	−0.7–0.7	11.7	1.3–22.1
9+	10.0	6.1–13.9	8.9	−0.2–17.9	−0.4	−2.5–1.8	−0.1	−9.0–8.8	5.0	3.8–6.2	15.7	8.1–23.3
Major ADG
0	−1.9	−3.7–−0.2	−10.0	−14.9–−5.0	−0.8	−1.9–0.3	−2.2	−6.6–2.2	0.0	−0.3–0.3	−10.0	−14.1–−6.0
1	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
2	2.6	−0.1–5.3	4.7	−1.6–11.0	−0.3	−1.7–1.1	4.8	−0.4–10.1	1.3	0.2–2.4	3.2	−2.4–8.8
3+	6.7	3.6–9.7	14.4	7.0–21.8	0.4	−1.2–2.0	10.4	4.0–16.8	3.3	1.8–4.9	17.3	10.1–24.5
Material deprivation quintile
1–least deprived	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
2	0.00	−2.2–2.2	−0.6	−6.9–5.8	0.5	−1.0–2.1	−1.9	−7.3–3.4	0.0	−0.2–0.2	−3.1	−8.0–1.8
3	−1.0	−3.5–1.6	−2.8	−8.8–3.2	−1.0	−2.5–0.4	−5.2	−10.1–−0.2	0.0	−0.2–0.2	−3.6	−8.7–1.5
4	−1.5	−4.0–0.9	−1.4	−7.8–4.9	−1.1	−2.7–0.5	−4.0	−9.6–1.6	0.0	−0.3–0.3	−4.6	−9.7–0.5
5–most deprived	−4.1	−6.4–−1.7	−12.3	−18.1–−6.5	−3.4	−5.0–−1.8	−11.3	−17.0–−5.6	0.0	−0.3–0.3	−2.4	−8.0–3.1
Rurality
Rural	−11.1	−13.4–−8.9	−4.5	−11.4–2.3	−5.8	−7.2–−4.4	−0.4	−6.8–5.9	0.0	−0.2–0.2	1.1	−6.0–8.3
Non-major urban	−5.1	−6.8–−3.4	−1.8	−6.5–2.9	−3.1	−4.2–1.9	−0.8	−4.8–3.1	0.0	−0.3–0.3	0.9	−3.4–5.2
Major urban	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
History of NMSC
Yes	16.2	12.9–19.4	35.3	28.3–42.3	4.4	2.5–6.2	11.5	5.7–17.3	14.0	12.0–16.0	44.0	38.4–49.5
No	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Established with dermatology
Yes	5.7	2.8–8.5	9.4	3.5–15.3	6.4	4.3–8.5	20.3	13.9–26.6	7.7	6.3–9.1	22.5	18.5–26.5
No	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Disease factors
Histology
Superficial spreading melanoma	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Nodular melanoma	−3.6	−5.7–−1.5	−10.7	−16.4–−5.1	−1.3	−2.7–0.1	−8.4	−13.5–−3.3	0.0	−0.4–0.4	−7.1	−12.6–−1.7
Lentigo maligna melanoma	4.5	1.1–7.9	10.4	2.4–18.4	2.5	0.4–4.7	3.2	−3.6–10.0	3.7	1.4–6.0	10.4	3.6–17.1
Acral lentiginous melanoma	8.4	−0.7–17.5	3.6	−16.4–23.6	4.3	−1.1–9.7	−1.1	−20.1–17.9	13.8	3.8–23.8	11.6	−0.3–23.5
Other	4.6	0.1–9.0	1.6	−9.9–13.2	1.8	−1.3–4.8	−3.6	−12.4–5.2	1.3	−0.6–3.3	7.1	−2.2–16.4
NOS	−2.5	−4.0–−0.9	7.5	2.5–12.5	0.5	−0.6–1.5	3.2	−1.0–7.4	0.0	−0.3–0.3	5.0	1.4–8.6
Location
Extremities	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Trunk	−0.8	−2.4–0.9	0.8	−4.0–5.5	−1.5	−2.6–−0.4	−4.1	−8.6–0.5	0.7	0.4–1.0	5.7	1.2–10.2
Face	11.4	8.4–14.4	11.4	4.5–18.3	3.5	1.7–5.4	1.8	−4.5–8.2	4.3	3.1–5.5	10.4	4.7–16.2
Scalp and neck	7.9	5.0–10.8	5.4	−2.6–13.5	2.4	0.3–4.5	−2.4	−10.0–5.1	4.7	3.2–6.2	12.9	6.1–19.7
NOS	−2.2	−5.5–1.0	15.1	3.2–27.0	−3.5	−5.7–−1.2	−3.3	−13.2–6.6	9.7	7.7–11.6	36.3	25.3–47.3

B is the difference (in days) from the reference category.

Bolded values signify statistical significance.

^aMutually adjusted for all variables in table.

ADG Adjusted Diagnostic Groups, NMSC Non-Melanoma Skin Cancer, NOS Not Otherwise Specified.

Patient characteristics

Sex

There was no difference in the length of the DI, PCI, or SCI by sex.

Age

There was little difference in the median DI, PCI, and SCI across age groups. However, there was an association between age and interval length when examining the 10% of patients who waited the longest. For example, the 10% of patients aged 20–45 who had the longest DI for their age group waited 16 days longer than the 10% of patients aged 56–65 who had the longest DI for their age category. The differences found in the DI were due to differences found mostly in the PCI, rather than the SCI. The 90^th percentile of the PCI was the longest for those aged 20–45 (16-days longer than those aged 56–65) and shortened with increasing age. In contrast, we did not see large differences in the SCI across age groups after adjusting for all other variables.

Comorbidities

The median and 90^th percentile of the DI increased with increasing comorbidity. Compared to those with 0–1 minor comorbidities, the median DI was 3 days longer for those with 2 minor comorbidities, and increased to 10 days longer for those with 9+ minor comorbidities. The presence of major comorbidities was similarly associated with longer DIs.

Material deprivation

Median DIs did not significantly differ across neighbourhood deprivation quintiles, however, the 10% of patients experiencing the longest DIs who lived in the least deprived neighbourhoods experienced DIs that were 12 days longer than those who lived in the most deprived neighbourhoods. This difference was also evident in the PCI, but not in the SCI.

Rurality

There were differences in the length of the DI according to rurality. The median DI for those living in rural areas was 11 days shorter than in those living in major urban areas, although the 90^th percentile did not differ. There were small differences (6-days) in the median PCI for those living in rural areas, and no differences in the SCI.

History of NMSC

The median and 90^th percentile of DI differed according to the presence of a previous NMSC diagnosis. Patients with a history of NMSC in the 5-years prior to their index contact date had a 16-day longer median, and a 35-day longer 90^th percentile of their DI compared to those without prior NMSC. Differences seen in the DI were due mainly to larger differences in the SCI (14 days and 44 days at the median and 90^th percentile, respectively).

Dermatology history

Patients previously established with dermatology experienced longer DIs, PCIs, and SCIs than those who did not have this prior history with a dermatologist. The median DI was 6-days longer, and the 90^th percentile was 9-days longer for those previously established with dermatology, compared to those without dermatology history and those previously established with dermatology had a 6-day and 8-day longer median PCI and SCI than those not previously established, and a 20-day and 23-day longer 90^th percentile of the PCI and SCI.

Disease characteristics

Histology

There was little difference in the median DI and PCI across histology categories. However, in the SCI, patients diagnosed with acral lentiginous melanoma had a 14-day longer median SCI than those diagnosed with superficial spreading melanoma. Variation across histology was evident at the 90^th percentile of the DI, PCI, and SCI. For instance, compared to patients with superficial spreading melanoma, patients diagnosed with nodular melanoma had shorter 90^th percentiles of the DI, PCI, and SCI (11, 8, and 7 days shorter, respectively), and patients with lentigo maligna melanoma had longer 90^th percentiles of the DI and SCI (10 days longer).

Anatomic location

There was little variation in the PCI by primary site of lesion, and significant variation in the length of the DI and SCI by the primary site of the lesion. Compared to patients with lesions diagnosed on the extremities, patients diagnosed with lesions on the face had 11-day longer median and 90^th percentile of DI, and 10-day longer 90^th percentile of SCI. Patients diagnosed with lesions on the scalp and neck had longer median DIs (8-days longer) and longer 90^th percentiles of the SCI (13-days longer) than patients with lesions on their extremities.

LHINs

We also found variations in the length of the DI, PCI, and SCI across LHINs. Median DI lengths ranged from 28 days to 45 days, median PCI lengths ranged from 15 days to 25 days, and median SCI lengths ranged from 1 day to 20 days. Adjusting for patient and disease characteristics did not explain all the differences that were seen, nor did the additional adjustment for county-level dermatologist density, suggesting that LHIN is independently associated with the length of the DI, PCI, and SCI (Table 4).

Table 4.

Diagnostic interval, primary care subinterval, and specialist care subinterval for local health integration networks (LHINs).

	Unadjusted		Case-mix adjusteda		Case-mixa and county-level dermatologist density adjusted
Characteristic	B	Confidence interval (95%)	B	Confidence interval (95%)	B	Confidence interval (95%)
Diagnostic interval
Intercept	37.0	35.1–39.9	30.8	26.5–35.1	36.1	31.1–41.1
LHIN
A	1.0	−2.7–4.7	−2.2	−5.7–1.3	−1.1	−4.7–2.5
B	−4.0	−7.6–−0.4	−2.4	−5.8–1.0	−1.7	−5.3–2.0
C	5.0	2.1–7.9	3.8	0.5–7.0	4.0	0.6–7.5
D	−2.6	−5.8–0.7	1.5	−1.8–4.8	−2.0	−6.1–2.0
E	−8.0	−10.7–−5.3	−4.0	−7.2–−0.7	−4.5	−7.9–−1.1
F	−2.0	−5.3–1.3	−2.3	−5.8–1.1	−2.3	−5.7–1.1
G	6.0	2.4–9.6	7.2	2.9–11.5	5.1	0.9–9.3
H	5.0	2.0–8.0	6.1	3.4–8.9	5.0	1.8–8.1
I	8.0	4.7–11.3	8.8	5.1–12.5	7.8	4.1–11.6
J	−9.0	−14.3–−3.7	−4.5	−9.7–0.6	−8.0	−14.0–−2.0
K	0.0	−4.5–4.5	0.3	−3.9–4.6	−0.4	−4.6–3.9
L	Ref		Ref		Ref
M	-10.0	−13.0–−7.0	−6.9	−10.1–−3.8	−8.4	−12.1–−4.7
N	−8.0	−11.6–−4.4	−3.0	−6.9–1.0	−4.0	−8.6–0.6
Primary care subinterval
Intercept	23.0	21.5–24.5	24.2	21.7–26.7	26.0	22.5–29.4
LHIN
A	0.0	−2.1–2.1	−1.8	−4.1–0.5	−1.4	−3.9–1.0
B	1.0	−1.8–3.8	0.7	−1.7–3.1	1.1	−1.4–3.6
C	1.0	−1.6–3.6	−0.6	−2.8–1.6	−0.4	−2.6–1.8
D	−2.0	−3.9–−0.1	0.9	−1.5–3.2	−0.2	−2.8–2.5
E	−4.0	−6.2–−1.8	−2.8	−4.9–−0.7	−3.3	−5.6–−0.9
F	2.0	−0.5–4.5	0.5	−1.7–2.7	0.7	−1.4–2.8
G	−4.0	−6.1–−1.9	−2.5	−4.6–−0.4	−3.4	−5.8–−0.9
H	1.0	−1.7–3.7	1.4	−0.6–3.5	1.1	−1.1–3.3
I	1.0	−1.2–3.2	0.4	−1.7–2.5	0.4	−1.9–2.7
J	−8.0	−13.7–−2.3	−7.3	−11.2–−3.5	−8.4	−12.5–−4.4
K	−1.0	−3.4–1.4	−0.6	−3.0–1.8	−0.6	−3.1–1.9
L	Ref		Ref		Ref
M	−3.0	−5.7–−0.3	−2.7	−5.3–−0.2	−3.7	−6.4–−1.0
N	−4.0	−6.9–−1.1	−2.2	−5.3–0.8	−2.9	−5.9–0.1
Specialist care subinterval
Intercept	5.0	3.5–6.5	1.0	0.7–1.32	1.0	0.3–1.7
LHIN
A	−1.0	−2.9–0.9	0.0	−0.5–0.5	0.0	−0.4–0.4
B	−4.0	−5.5–−2.5	0.0	−0.2–0.2	0.0	−0.2–0.2
C	4.0	2.1–5.9	3.0	1.9–4.1	3.0	1.8–4.2
D	3.0	1.5–4.5	2.6	1.5–3.7	2.6	1.1–4.1
E	−3.0	−4.8–−1.2	0.0	−0.2–0.2	0.0	−0.3–0.3
F	−3.0	−5.0–−1.0	0.0	−0.2–0.2	0.0	−0.2–0.2
G	15.0	12.8–17.2	12.6	11.2–14.0	12.6	11.0–14.2
H	2.0	−0.1–4.1	1.0	0.3–1.7	1.0	0.3–1.7
I	9.0	6.8–11.2	9.8	8.7–10.9	9.8	8.4–11.2
J	9.0	5.8–12.2	8.0	5.6–10.4	8.0	5.5–10.5
K	2.0	−0.4–4.4	0.6	−0.5–1.7	0.6	−0.6–1.8
L	Ref		Ref		Ref
M	−4.0	−5.5–−2.5	0.0	−0.2–0.2	0.0	−0.4–0.4
N	−1.0	−2.6–0.6	0.0	−0.4–0.4	0.0	−0.4–0.4

Bolded values signify statistical significance.

Results from quantile regression at the 50^th percentile. Unadjusted, case-mix adjusted^a, and case-mix^a + county-level dermatologist density- adjusted. B is the difference (in days) from the reference LHIN.

^aCase-mix: sex, age, minor ADG, major ADG, deprivation quintile, rurality, histology, location, history of non-melanoma skin cancer, previously established with dermatology.

LHIN Local Health Integration Network, ADG Adjusted Diagnostic Groups.

Sensitivity analyses

We conducted a sensitivity analysis on the data from the subset of patients with staging information, to examine stage as a possible explanatory variable for our findings. This adjustment did not significantly alter our estimates except for histology, where differences in DI, PCI, and SCI were no longer significantly different across histological subtypes (Supplementary Appendix E). Of note, adjusting for stage also did not significantly affect the LHIN parameter estimates beyond adjustment for patient- and disease-factors (Supplementary Appendix E).

Discussion

This study provides population-level estimates of the DI experienced by melanoma patients covered by a universal health care system. The median DI we observed, of 36 days, exceeded the provincial recommended limit of 28 days, and there was substantial regional variation in the DI. Variation in the DI and our subintervals suggest uneven access to timely primary care and specialist care. We also identified patient, disease, and system factors that were associated with this variation. For example, older patients had a shorter DI, driven largely by the length of the SCI. Those with comorbidities tended to have a longer DI, driven by differences in both the PCI and SCI. The DI was longer for scalp and neck melanomas, driven by the SCI. The DI varied between health regions, driven more by the SCI than PCI.

Half of the patients in our cohort waited longer than 36 days, and 25% waited longer than 85 days to be diagnosed with melanoma. Our estimates are longer than most previous studies measuring the DI [16, 18, 19, 37–39]. Differences in estimates may be due to the inconsistencies in definition of the DI, differences in the populations and health systems examined, and differences in the method of ascertaining the index date. Our results are most similar to a population-based study conducted by Torring [38], which similarly defined the DI according to the Aarhus statement and instructed physicians to use medical records to assign the index date. This study was conducted in Aarhus County, Denmark, which has a similar healthcare system to Ontario, where GPs act as gatekeepers to the rest of the universal healthcare system. Our results are also similar to the only other study conducted in Canada, which reported a median physician delay of 1 month, with 65% of patients waiting 0–3 months and 15% of patients waiting longer than 6 months from initial presentation to excision [21].

Not only was the provincial DI long, but we also saw regional variation in the length of the DI. We found differences in the length of the DI according to health region of residence (LHIN), with two LHINs experiencing adjusted median DIs that are one week longer than the reference LHIN, and one LHIN experiencing an adjusted median DI that is one-week shorter. Three LHINs experienced median SCIs that were more than one week longer than the reference LHIN. There was no clear relationship between the length of the PCI and SCI in a LHIN, and controlling for differences in patient and disease factors, dermatologist supply across the LHINs, and stage did not explain the observed differences. This suggests that other factors, such as access and quality of care, may differ across the province. Indeed, we previously found variability in the risk of advanced thickness melanoma across LHINs [26], further suggesting that future research should focus on differences in care across the province.

Younger patients have a significantly longer 90^th percentile of DIs than older patients. The finding of longer PCIs in younger patients and shorter PCIs in older patients is consistent with previous research, which found that younger cancer patients were more likely to have multiple pre-referral consultations, compared to older patients [40]. However, it seems from our results that once a patient consults with a specialist, the specialist recognizes the melanoma and promptly performs a biopsy, regardless of age.

Increasing comorbidity was associated with increasing DI length, with larger differences at the 90^th percentile. One proposed explanation for this finding is that the comorbid conditions present symptoms which may mask the cancer symptoms, leading to misdiagnoses [41]. However, 82% of patients in the subset with stage data had early-stage disease. Since early melanoma does not present with symptoms that may be masked by comorbid conditions, an alternative explanation for our findings is that the comorbid conditions may take precedence in a patient’s care [41], patients were on blood thinners that needed to be stopped prior to biopsy, or patients may be too ill to attend an appointment at the recommended time, thus delaying the specialist visit and/or biopsy.

After adjusting for patient and disease factors, those living in the most deprived neighbourhoods, and those living in rural areas had shorter DIs overall. These differences were largely driven by differences in the PCI. A shorter DI in those living in deprived neighbourhoods may be partly explained by their diagnostic route. We have shown that a higher proportion of patients living in deprived neighbourhoods and rural areas are diagnosed in primary care (unpublished). Because these patients do not have to wait to see a specialist, they are diagnosed faster than those who are referred to a specialist for a biopsy.

Those diagnosed with nodular melanomas had the shortest DIs overall, with adjusted 90^th percentile that was 11-days shorter than those diagnosed with superficial spreading melanomas, and those diagnosed with lentigo maligna melanomas experienced the longest DIs, with adjusted 90^th percentile that was 10-days longer than those diagnosed with superficial spreading melanoma. Those diagnosed with melanomas on the face experienced longer DIs overall. Previous studies have shown similar patterns [16, 42]. In our previous study [26], we found that nodular melanomas are often thicker and ulcerated when diagnosed, suggesting that they are more often symptomatic, garnering more attention from the attending physician. Lentigo maligna melanomas are often diagnosed on the face and are more common in the elderly. Due to their conspicuous location, physicians may be less willing to perform a cosmetically damaging biopsy, and attempt to either perform other, less-damaging procedures, or refer to a specialist to perform the excision, which may result in longer waits [43].

Those with a history of NMSC and those previously established with a dermatologist had longer DIs. We hypothesised a priori that those with a history of NMSC and those previously established with a dermatologist would experience shorter intervals due to increased surveillance. There are two possible explanations for these findings. One is misclassification of the index date for these patients, where we captured healthcare visits related to the NMSC or a previous dermatologic problem and misattributed them to the melanoma, resulting in an artificially long DI. The chance of this occurring is quantified by 1 minus our signal strength estimates, which ranged from 80% to 96% (Supplementary Appendix B). We limited the encounters that could start the diagnostic interval to those deemed likely to be related to melanoma, rather than other benign skin conditions. To mitigate the impact of potential misclassification on interpretation of results, we adjusted for the longer intervals for dermatology and NMSC history in our models. The other possibility is that our findings of a prolonged diagnostic period reflect an important phenomenon for melanoma diagnosis that should be explored and addressed. Prolonged DIs in patients with a previous NMSC or previously established with a dermatologist could be the result of added time waiting for specialist assessment of lesions detected by patients or primary care following the last specialist visit, or complacency on the part of the patient. If a patient feels as though they are being adequately followed, they may not see their lesion as requiring urgent attention, or they (or their physician) may misattribute the lesion for another NMSC or other benign condition, rather than a more serious melanoma. Further research should examine patients with a history of NMSC and those previously established with a dermatologist to determine the causes of a long DI.

This study has several strengths. Our melanoma cases came from a population-based cancer registry, with melanoma capture rates of 95% or more for melanoma diagnosed in 2007–2009 [25]. This high level of coverage is important as melanoma can be diagnosed and treated in a variety of care settings including physician offices, outpatient clinics, and hospitals. Many previous studies have been conducted on patients referred to specialised dermatology, plastic surgery, or melanoma clinics, and their results do not represent the pattern of care that most patients receive. [15–17, 19, 37] Our high rate of coverage ensures the representativeness of our cohort to all melanoma patients in Ontario and ensures that the patterns we observed are true representations of the care melanoma patients are receiving, at a population-level. While the magnitude of our estimates may be specific to Ontario patients, we believe the patterns we observed are generalisable to regions with a similar healthcare delivery model.

We used an empirical approach to identifying relevant encounters and statistical control charts to determine encounter lookback periods and hence the index date of the diagnostic interval for each patient. We also determined lookback periods for 3 separate eras to account for changes in practice. By using an empirical approach to calculate the DI, we improved on previous studies which used arbitrary lookback periods and encounter types.

We separately examined the primary care and specialist care subintervals. We found that different factors were important determinants of the length of the DI, PCI, and SCI. The PCI and SCI also relate to different aspects of the healthcare system, making our findings more useable and targetable, compared to studies only examining the overall DI. Future research could examine the differential importance of the DI, PCI, and SCI on outcomes.

Finally, we performed univariate and multivariate quantile regression models to examine the effects of factors on the 50^th and 90^th percentiles of the DI, PCI, and SCI distributions. Studying associations at the 50^th percentile provides information about differences in the majority experience while examining the 90^th percentile allowed us to identify patients and factors associated with extreme waits.

There are also several limitations to this study. The determination of the DI relied on physician billing claims. If a patient presents multiple ailments to their physician, a diagnosis code relating to the melanoma may not be recorded, and we would not identify that visit as relating to the melanoma. This could lead to an underestimate of the DI.

There has been no independent validation of our method for assessing the DI. Such a validation would require accessing multiple physician records that may or may not be more accurate and complete than the administrative data.

We used patients’ own healthcare encounters to create the list of melanoma-related encounters and compared frequencies to those from a control-period of 12–15 months prior to diagnosis, which we felt adequately represented a background rate of these encounters for this patient population. If the index date occurred more than 12-months prior to diagnosis, it is possible that we underestimated the length of a patient’s DI. We used control-charts to assign lookback periods for each encounter and for each era of patients, and assessed the signal strength at the lookback period, which is the probability that the encounter represented diagnostic activity of melanoma, rather than a background rate of encounter. We sought to achieve a minimum of 80% signal strength for the encounter category to be considered ‘melanoma-related’, so it is possible that some of the encounters we assigned as ‘melanoma-related’ were not related to the melanoma diagnosis.

We did not examine the relationship between the DI and mortality as there is not a clear understanding of the shape of the relationship between diagnostic delay and death. The natural history of the disease, as well as both patient- and treatment- factors can all contribute to mortality. We also acknowledge that many patients had NOS histology, though this is common in population-level registries [44, 45].

Finally, we were unable to control for stage and race in our multivariate regression models. Stage capture rates are low in the OCR for most disease sites, and information on race is not routinely available in Ontario’s health administrative databases. Although we were able to supplement stage capture for a random subset of our cohort using primary data collection of abstracted pathology reports, we were only able to ascertain stage for 55% of our cohort. We performed sensitivity analyses on those with stage information, and adjusting for stage did not alter our conclusions, except for histology, where stage explained the associations seen. (Supplementary Appendix E).

Conclusions

This study found variability in the melanoma DI, with half of the patients waiting longer than 36 days, and 10% of patients waiting longer than 142 days for diagnosis, which is much longer than the provincial guideline of 28 days for biopsy for all suspicious lesions [46]. We found various patient-, disease-, and system-level factors that were determinants of a long DI, PCI, and SCI, including age and comorbidity. Patients living in the most deprived neighbourhoods and those living in rural areas had the shortest DIs, due to shorter PCI as opposed to shorter SCIs. The finding of different associations in the primary and specialist care intervals highlight the importance of separately examining the sub-intervals, as they relate to different aspects of the healthcare system. Finally, of note is a variability in the DI across the health regions, suggesting the need for system-level quality improvement. From a public health standpoint, strategies should aim to reduce both the median DI and the 90^th percentile of the DI; our results highlight groups more likely to experience long DIs and groups where efforts should focus.

Author contributions

MEM had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. TPH and PAG contributed equally to this work as co-senior authors. MEM: conceptualisation, design, statistical analyses, interpretation, original draft, revision, review. YA: clinical expertise, interpretation, review. HL: clinical expertise, interpretation, review. NJLH: clinical expertise, interpretation, review. PN: cohort creation, data processing, review. FCW: clinical expertise, interpretation, review. TPH: conceptualisation, design, supervision, interpretation, revision. PAG: conceptualisation, design, supervision, interpretation, revision.

Funding

The authors received no specific funding for this work. MEM is supported by a Canadian Institutes of Health Research Doctoral Award.

Data availability

The dataset from this study is held securely in coded form at ICES. While legal data sharing agreements between ICES and data providers (e.g., healthcare organisations and government) prohibit ICES from making the dataset publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS (email: das@ices.on.ca).

Competing interests

The authors declare no competing interests.

Ethics approval

This project has been approved by the Health Sciences Research Ethics Board at Queen’s University. Participant consent has been waived as the research falls under Article 5.5A of the Tri-Council Policy Statement 2 (TCPS 2).

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-023-02518-1.